Embed Size (px)
Citation preview
AD-Ai32 989 AN INVESTIGATION INTO THE COUPLING OF INTERACTIVE AND 1/3BATCH NETWORK SERVICES IN COINS(U) NAVAL POSTGRADUATE
UNCA17SIFEDSCHOOL MONTEREY CA J 8 KIM JUN 83 FG92 N
d1.
*16
111111.25 12*fl~ U 12
MICROOPY RSOLIONTETCHRNAINA URA F INDRS-%-
. .. . .. . . .
00
NAVAL POSTGRADUATE SCHOOLCID Monterey, California
THESIS A
AN INVESTIGATION INTO THE COUPLINGOF INTERACTIVE AND BATCH NETWORK SERVICES .-
IN COINS
by
Joanne Bong Soon Kim
June 1983
CDTrhesis Advisor: N.F. Schneidewind
-~j
t1.Approved for public release; distribution unlimited.
* ~~SECURITY CLASSIFICATION OP THIS PAGE (Urn., oe asmloo_________
PAGE READ INSTRUCTIONSREPORT DCUMENTATION PAG RFORE COMPLETING FORM
I. WIPQ1RTRUNN. GOVT ACCESSION No 3. RECIPIaNT'S CATALOG NUMBER
4. TITLE (OW &I*) S TYPE Of REPORT & PERIOD COVERED
*An Investigation Into The Coupling of Master's Thesis;Interactive and Batch Network Services June 1983in COINS 41. PERFORMING ONG. REPORT NUNSER
7- AUTHOR(A) 11- CONTRACT OR GRANT NUUSER(o)
Joanne Bong Soon Kim
SPERFORING ORGANIZATION NAME AND ADDREISS W0 PROGRAM ELZMENT. PROJECT. TASK* AREA & WOPIC UNIT NUMISERS
Naval Postgraduate SchoolMonterey, California 93940
I I. CONTROLLING OFFICE NAME AND ADDRESS 12. REPORT DATE
*Naval Postgraduate School June 1983Monterey, California 93940 13. NUMSER OF PAGES
___ ___ ___ ___ __ ___ ___ ___ __ ___ ___ ___ __ 194
14- MONITORING AGENCY NAME 6 AODRESS(H d~throut krn Controllng OfiSce) IS. SECURITY CLASS. (of this report)
Unclassified5se. DECLASSIFICATION' OWNGRADING
SCHEOuLE
16. 0STRIW@VTION STATEMENT (of tis Aen)t
Approved for public release; distribution unlimited.
17. 0DSTA111UTION STATEMENT (1of Mebet 0aot 8td In DWoob It diff.,mt bass Repot)
IS. SIJPP14.MSN9TARV NOTES
IS. KEy WRNDS (Coalkoe aooo sidesof Necesar aowpd idmlS* by block Memnber)
NetworksInformat ion-sharingBatch and Interactive CouplingSimulation
W 4 AISTRACT (Cmffirne FOPWO reersldo it necessary sad ides.Qei by block numb.)
Networks were conceived in the 19501s, born in the 1960's* and grew up in the 1970's. Today they consti6tute a technology
with applications in a myriad of disciplines. Informationsharing has been one of the areas greatly aided by computernetworks. The Community On-Line Inteliigence System (COINS)is an information sharing network in the U.S. intelligencecommunity. COINS offers batch and interactive services which
DO 0 1073 EDTION OF I Noy es is oSSLETE
S/N 0102- LP. 014- 6601 SECURITY CLASSIFICATION OF T1418 PAGE (When Dote Bate,.e
IS MV CLASUPICATION OF THUS PAOS (m DOS Bmftt
are separate and independent of each other. The informationacquisition process has elements of interactive and batch.The design of an information sharing network should provideI.the foundation to accommodate this two-phased activity.- Thisthesis introduces the concept of collaboration between theseautonomous network services, proposes a re-allocation ofnetwork capacity in COINS and examines how this new schemecan improve performance and efficiency from a user andmanagerial perspective.
SIN~A cr- 012-L- 1-60
2~
SICURITY CLASSIFICATIO04 OP TWIS PAGWfUS. DOP Dmeetd)
Approved for public release; distribution unlimits-A."
An Investigation Into The Couplingof Interactive an Batch Network Servicesfn COINS :>
by
Jcanne Bong Soon Kim
B.A., University of Hawaii, 1966
Submitted in partial fulfillment of therequirements for the degree of
MASTER OF SCIZNCE IN SYSTEMS TECHNOLOGY(Command, Control and Communications)
from the
NAVAL POSTGRADUATE SCHOOLJune 1983
Author: 44,
Approved by: -Z"-Thesis Advisor
Second Reader
Chairman, com d nrland
ommunications Academic GrCup
Academic Dean
3. °-
& BSTR&CT
Netvwcks were ccnceived in the 1950's, born in the
19601s and grew up in the 19709s. To4ay they constitute a
technology with applications in a myriad of disciplines.
Informaticn sharing has been one of the areas greatly aided• "by computer networks. The Community On-Line Intelligence
System (COINS) is an information sharing network in the U.S.
intelligence community. CCINS offers batch and interactiveservices which are separate and independant of each other.The inforzation acquisition process has elements of interac-
tive and batch. The design of an information sharingnetwork should provide the foundation to accommodate thistwo-phased activity. This thesis introduces the concept of
collatoraticn between these autonomous network services,proposes a re-allocation of network capacity in CCINS andexamines how this new scheme can i2prove performance and
efficiency from a user and managerial perspective.
g4
5Z.
IABLE CF CONTENTS
" 'I. I IDCCIUCTION .................. 11I.o THE CaINS NETWORK . . . . . . .. 13
A. TB!Cciis NE . . . . . . . . . . . . . . . .. . 13
E. CURRENT SYSTEM . . . . . . . . . . . . . . . . 16
C. FUTURE PLANS .. ........... ... . 16
III. 11TUCRK CONFIGORATION . . . . . . . . . . . . . . 19
A. THE SUBNET ENVIRONMENT . . . . . . . . . . . . 19
B. HARDWARE NVIRONENT . ........... 19
C. SOFTWARE ENVIRONMENT . . . . . . . . . . . . . 20
IV. USE SERVICES ............... 2
A. BATCH SERVICES . . . . . . . . . . .... 22
B. INTERACTIVE SERVICES . . . . . . . . . . . . . 22
1. Intelligence Databases . . . . . . . . . . 22
2. User-Support Databases . 23
3. Managerial and Administrative Databases 25
V. NEIHCDOLOGY .................. 26
VI. USER PROFILE INALYSIS . . . . . . . . . . . . . * 28
A. GENERAL wee . ease .. . . . .. e e a . 28"0 B. MODEL DESCRIPTION. . . .. .. . ... . .. 31
C. TIME-LIVE INSPECTION . . . . . . . . . . . . . 341
VII. AN ALTERNATIVE APPROACH:COUPLING OF INTERACTIVE
AND EATCH a e e a e a . e e o a 38
VIII. THE NETWORK SERVICES MODEL . . . . . . . . . . . . 42
--* A. ESIGN GOILS AND CONSIDERATIONS ...... 42
B. HARDWARE AND SOFTWARE ENVIRONMENT . 46
C. FUNCTIONAL DESIGN . . . . . . . . . . . . . . 47
1. Customer Arrival . . . . . . . . . . . . . 47
5
"K Q . .o. -,.- ... " ; .. - .. ' . . .• . . . . .•. . . . ... . . . . .. . . . -
2. Resour.ce Selectiono . . ........ 48
3. Service Profile . . . . . . . . . . . . 48
4. High-Speed Queueing Discipline . . . ; . 50
D. INTERNAL DESIGN . . . . . . . . . . . . e 51
1. Overall Structure ............ 51
2. Event Handling . . . . . . . . . . . . . . 56
3. Data Structures . . . . . . . . . . . . . 60
P. TRELIKINAFY RESULTS . . . . . . . . . . . . . 63
F. MODEL VALIDATION . . . . . . . . 69
G. MODEL APPLICATIONS .............. 70
IX. CCIPARETIVE ANALYSIS . . . . . . . . . . . . . . . 71
I. CCECLUSIONS AND RECOMHENDATIONS . . . . . . . . . 80
A. SUMMARY . . .. . . .. . . . . . . . . .. . 80
1. Current Environment ........... 80
2. Pcpulation Growth ............ 80
3. Data Transfer Growth . . . . . . . . . . . 81
4. Review . . . . . . . . . . . . . . . . . . 82
B. CONCLUSIONS ................. 82
C. RECOMMENDITIONS . . . . . . . . . . . . . . . 83
APPENDIX A: EMPIRICAL DATA ANALYSIS .8.4........
APPENDIX E: OPERATING INSTRUCTIONS FOR INS MODEL .... 89
A. HOST CHARACTERISTICS . . . . . . . . . . . . . 89
S. NETWORK CCNFIGURATION . . . . . . . . . . . . 90
C. AS AND SEUVER-TAS CHARACTERISTICS ...... 91
APPENDIX C: TABLES CF RESULTS . . . . . . . . . . . . . 93
APPENDIX 1: SYSTEM TIME AND EXPECTED LOSS CHARTS . . . . 96
APPENEIX I: FIVE TO EIGHT TAS CONFIGURATION ..... 117
APPENDIX F: EVENT LCGIC DIAGRAMS . . . . . . . . . . . 120
6
&PPIUCII G: INlS PROGEAM LISTING . . . . . . . . . . . 132
INITI S2IAEIIJTIOILLST.IS..T ... .. .... . . 193
LIST OF TABLES
I. TAS operating characteristics . . . . . . . . . . 63
II. Increasing arrival rates . . . . . . . . . . . . . 65
II. Increasing data transfer sizes . . . . . . . . . . 67
IV. flcdel validation results . . . . . . . . . . . . . 69
V. Adding a TAS tc the netork ........... 72
VI. Propcrtion of interactive use . . . . . . . . . . 75
VII. Ranges of system times and expected loss . . . . . 76
VIII. Ranges of proEcrtion of interactive use . . . . . 77
IX. Condensed Comparison Chart . . . . . . . . . . . . 82
X. SCLIS interactive-only time analysis . . . . . . . 8
XI. SCLIS interactive time analysis . . ....... 85
XII. SCIIS data transfer time analysis . . . . . . . . 86
XIII. TASI inter-arrival time analysis . . . . . . . . . 86
XIV. UZS2 inter-arrival time analysis . . . . . . . . . 87
IV. TAS3 inter-arrival time analysis . . . . . . . . . 87
XVI. !AS4 inte--arrival time analysis . . . . . . . . . 88
XVII. Five TkS configuration .............. 93
XVIII. Six TS configuration . . . . . . . . . . . 94
XIX. Seven TAS configuration . . . . . . . . . . . . . 94
XX. Eight TAS configuration . . . . . . . . . . . . . 95
XI1. 5 UIS, 89,018 characters: expected transfer
amcunt 97.920.characters:...117
XXII. 5 lAS, 97,920 characters: expected transfer
acunt 173 ..71.characters:..x118
XXIII. 5 lAS, 107,712 characters: expected transfer
48
ascunt 118XXIV. 5 7AS, 173,471 characters: expected transfer
amount . . . . . . . . . .. 119
XXV. 5 7AS, 190,818 characters: expected transfer
amount .. . . . . . . . . . . . . .. . 119
. - . • o . . . ..8
LIST OF FIGURES
2.1 Criginal COINS configuration . . . . . . . . . . 114
2.2 Current COIlS-I configuration . . . . . . . . . 17
6.1 Cne populaticn source to SOLIS . . . . . . . . . 32
* 6.2 FcUr populaticn sources to SOLIS . . . . . . . . 33
6.3 2-Stage service facility . . . . . . . . . . . . 35
6.4 Time-Line diagram of 1 retrieval session . . . . 36
6.5 Tize-line of 3 retrieval sessions . . . . . . . 37
7.1 SOLIS as a 2-node tandem network ........ 39
7.2 Time-line with 1 data and 2 interactive paths . 408.1 Initial State ................. 53
8.2 State 2 . .. .. . .. . . . . . . . . . . . . 53
8.3 State 3 . .... .. ... . .. .. . .... 54
8.4 State 4 . . . . . . . . . . . . . . . . . . . . 54
8.5 State 5 . . . . . . . . . . . . .. . . . .. . 55
8.6 State 6 . . . . . . . . . . . ..... .... 55
8.7 State 7 . . . . . . . . . . 0 0 56
8.8 System times with increasing arrival rates . . . 66
8.9 System times with increasing transfer sizes . . 68
9.1 System times with increasing TASs ....... 73
9.2 Expected loss rate . . . . . .......... 74
D.1 system times with 89,018 characters . . . . . . 97
D.2 Expected loss with 89,018 characters ...... 98
D.3 System times with 97,920 characters . . . . . . 99
D.4 Expected loss with 97,920 characters . . . . . 130
0.5 System times with 107,712 characters . . . . . 101
D.6 Expected loss with 107,712 characters . . . . 102
D.7 System times with 118,483 characters . . . . . 103
D.8 Expected loss with 118,1183 characters . . . . 10 4
D.9 System times with 130,331 characters . . . . . 105
D.10 Expected loss with 130,331 characters 0 a 0 0 106
9
* * * ~ ***.*'..K---:-J..:..w:;.% Ii'i-. . -.
1i System times with 143,364 characters ,. .. 107
D. 12 Expected loss with 143,364 characters ... 108
D.13 System times with 157,701 characters . . . . . 109
D.14 Expected loss with 157,701 characters .... 110D.15 Systm ties ith 173,471 characters . . . . .11
D.16 Expected loss with 173,471 characters .... 112
D.17 System times with 190,8 18 characters . .113
D.18 Expected loss with 190,818 characters . 0 . 114
D.19 System times with 209,900 characters . . ... 115
D.20 Expected loss with 209,900 characters . ... 116
F.1 MAIN oR.I . e . . . * * . * e . e . . e e . . . 122
P.3 A AHEPRRIVAL . . . . . . . . . . . . . . . . . 123
F.5 UC.IERIVAL . . . . . . . . . . . . . . . . . . 125
F.6 SC.ARRIVkL . . . . . . . . . . . . . . . . . . 126
P.7 SC.DRPkRT .................. 127
P.8 SC.DZPART (ccntinuaticn 1) . . . . . . . . . . 128
P.9 SC.DEPART (ccntinuati4on 2) . . . . . . . . . . 129
F.10 ITHDEP&ET . . . . . . . . . . . . . . . . . . 130[.1.1SIG 3
10
.0
I. _! 10.QUCTIO!
In recent years* we have witnessed changes in informa-
tion searching and sharing practices. With the dramatic
decrease in digital technology and the concomitant advance-
ments in computing and communications, we have seen the
birth of the new technology cf networking. Like any tool,
which is viewed as a solution to a problem, so too
netwcrking and in its many forms has been brought to bear on
a variety of problems [Ref. 1], [Ref. 2]. Information
sharing has been one cf the areas greatly aided by this new
technology. It is now possible to have real-time interac-
tive access to massive amounts of information around the
globe at the touch of one's finger-tips. There are numerous
informaticn sharing networks in private industry, the
Department of Defense (DoD) and other government agencies.
• One of these is the Community On Line Intelligenc. System
(COINS) which interccnnects on-line information storage and
retrieval systems iccated at several locations within the
U.S. Intelligence Ccmmunity. COINS provides world-wide
access tc these information resources.
There was a point when the general belief was that all
network access of the future would be interactive with a
demise of batch processing. However, this view has been
ameloriated after a close inspection of the user needs.Batch prccessing is still very useful and desirable. Inmany cases, a batch facility can enhance the analytic use of
interactive services. COINS has both batch and interactive
network facilities, but they are separate and independent.
In each domain, we can envisage users attempting to use
each facility to do tcth interactive and batch work. What
is called for is some type of coupling of interactive and
. .. 11
a e " " .° . .'.. ." . - .L°.j• .1"L.J.2-i1 -. • . . .... . . .. .
batch network capabilities which matches users' needs. The
purpcse of this thesis is tc introduce the concept cf cclla-
boration between otherwise autonomous operations and to
study a scheme reflecting this synergetic notion. We will
examine what effects this has on performance and efficiercy
from a user and manager perspective. We have designed and
implemented a computer simulation of the flow of user
requests to the interactive facility in COINS to help study
the merits cf the twc approaches.
This thesis is organized as follows:
1. description of the COINS network, its architecture and
its current implementation techniques for interactive
information sharing;
2. discussicn of the evaluation criteria for network
Ferfcrmance;
3. presentation cf an alternate proposal with discussion
of how this new scheme is likely to improve interac-
tive informaticn sharing;
4. description of the interactive network services simu-
lation model;
5. discussion of preliminary results using the 3odel;
6. discussion of the simulation model's applicabili.y in
evaluating an alternative capacity allocation strat.gy
as COINS grows; and
7. ccnclusions and recommendations.
12
:;':::::,:'::' .......... ':' ..........;:.,:..........--:"-.--: - ,... . . ... . .. . . ......
1i. ZU GOINS P-22
1. BZCECOIUND
The Ccmunity On line Intelligence System was establised
on the recommendation of the President's Foreign
Intelligence Board (EPIB) in June 1965 to improve informa-
tion handling methods. The implementation plan called for a
star-configured netwcrk to provide connectivity amcng the
intelligence data prccessors. The concept was to permit an
analyst sitting at his local terminal to access infcrmationeither at his host processor or at a remote centralprocessing unit (cpu). The participants were the Central
Intelligence Agency (CIA), the Defense Intelligence Agency
(DIA), the National Security Agency (NSA), the Nat'onalPhotographic Interpretation Center (NPIC) , the StateDepartment, and the National Indications Center (NIC). The
store-and-forward message switching node was physically
Iccated at DIA. igure 2.1 shows the original COINS
configuraticn. Implicit in this concept was the requirementfor an intelligence crganization to have a cpu connected to
the COINS-switch to access information in COINS. The State
Department and NIC did not have cpus in COINS. Hence access
for tkese two organizations and any others that did not have
cpus was by prccuring a terminal from one of the hcst
processors in the network.Each of the nodes offered the same batch query and
retrieval services to the network as they did to their local
users. Users wculd submit their network queries at theirlocal terminals and some time later would receive their
responses. Depending on the data manipulation tools of the
host, the responses would range from simple data record
13
1'
terminals terminals
I! I
.. ... cp@
-i IDl IkNPIC
jel COINS SWITCH l
11
-- Icpu Cpu
[" ~~~I *... ....
I.I1oig ra.1ls gnl oi confgurainl
.
_. ,. . ....... . ... ,..... .. ....%* * * ... * . . . . . . .. . .:. -'-, ".. .',-.. , --* -2.' .. '-- . '. . '. -',-*','*.*.' - ... - . -. . -... '- . ... .. " .. . . . . . " •
listings to some statistical summaries of numeric informa-
tion.
Over the years CCINS access and participation expanded
with a netting of several military commands under the
management cf DIA. It was called the Intelligence Data
Handling S7stem, Ccmmunications (IDHSC). Its form wassimply the extension of the star-configuration.
By 1975, significant developmeats in teleprocessing
provided the necessary impetus for COINS to move to the next
chapter in networking with the introduction of new network
services. COINS assimilated into its architecturs the
following:
1. the Advanced Research Projects Agency's (ARPA)
networking technology of packet-switching;
2. the Sigint On Line Information System (SOLIS), an
interactive, full-text retrieval system;
3. a user-Terminal Access System (TAS); and
4. Front-Ead processors which connect the database cpu's
tc the new networking technology.
The concept of a TAS was necessitated by a grcwjng
number cf intelligence organizations without Cpus that
wanted access to COINS. This requirement was further rein-
forced by the private sector idea of relieving the database
hosts of terminal handling functions and putting all user
interfaces cn a separate facility. The TAS provides beth
batch query services to the batch hosts and interactive
query services tc the interactive host. with the adoption
of this new technology and services, the network was named
COINS-II. The IDHSC component of COINS kept the star-
configuraticn.In 1976, COINS-II undertook an internetting experiment
with the ARPANET, installing a TAS in Hawaii. As a result
the Pacific Command (PACO) now has secure inte:act1ve
access tc the full-text retrieval system.
15
By 1S80, COINS-II introduced a new kind of TAS. ih Ile
the original TAS is a pure user which does not offer any
databases tc the network, this new TAS has on-line user
support functions and network management information. This
type cf TAS will be denoted as a server-TAS.
B. CRRINT SYSTUR
Figure 2.2 is a picture of COINS-II today. Currently
there are two ZASs, two server-TASs, one interactive, full-
text retrieval host, and five batch retrieval hosts. The
user TASs are called TAS and AKU. The first TAS retained
the name lAS. The server-TASs are called NSH (Netwcrk
Service Hcst) and TIP (Transfer Research Facility). With
the COINS/FMO develcping a family of TASs, the NSH has
evolved intc a MASTEB-TAS, similar to the concept of the
MASTER-IIRE in AREA technology whereby sof4tware releases and
remote debugging are done. NSH also supports a small cadre
of operaticnal users. TRF is the COINSts developmental
* facility where research ideas can be developed and tested in
an operational ervircnment. The user-support system resides
on TRE. No0 intelligence analysts are supported by TRF. The
database losts have Front-Ends (PEs) to connect them tc the
communicaticns network. Each retrieval system has its cwn
language and each data file has its own coding schemes. As
part of the ARPANET technology, there is a Network Ccnr-rol
Computer 4NCQ for the communications network mcnitorinq and
management.
C. PUTUIR HAINS
Vith respect to network growth, the COINS/P150 antici-
pates two more interactive server-hosts and four more TASs
by 1985 [Ref. 3]. In the area of network services, the
COINS/EMC has a joint effort w~th the Department of the Army
16
- r~ 47 77 W -. -W - .;. - --- V -; -. C 4- t-a a
BATCH BITCH
MCC TAS ISH TRF HOSTI SOLIS HOST2
subnetvork I
'a !
---. -----------
I azpanet I
a. IAU swi-ichl
terminal BAICH EATIH terminalaccess HOST3 HCST4 access isystem system switch2
I (te-minal accesssystem)
BAICH BAICHHOST5 HOST6
igure 2.2 Current COINS-I configuration.
17
in data fusion analysis [Ref. 4]. For temutipiq quer
language problem, COINS is continuing in effort called.ADAPT
which is a network larguage that users would employ. ADAPT
would make the appropriate transforms to the target
languages. The COINS organization is working with the
Center fcr Computer Security at NSA on the multi-level
security problems in networks and in inter-networks. COINS
also has several AEPA-sponsored efforts in ths area of
human-factors engineering for the network user. Here the
concern is with the work environment and the develcpment of
an intelligence analyst work-station of the future (Ref. 5].
18
a.2.
Netvcrk configuration is a combination of commurica-
tions, hardware and software. These components are
discussed in this charter.
A. TB! SUBNE RIVIROUUET
When the term "sub-network', is used below, it will mean
the communications technology supporting host connectivity.
The sub-network configuration of COINS-II is the packet-switching technology of ARPANET. The six IMPs (Interface
Message Frocessors) are distributed over five sites in the
Washingtcn, D.C. area and are connected by 56 kilobit/second
phone lines. The IMPs are a mix of Honeywell- 316s and
C-30s. With the use of a special gateway (GW) and a pair
cf private line interfaces (PLIs), COINS-II is internetted
Swith the ARPANET, resulting in connectivity to Hawaii.
There are 15 hosts in COINS. There are seven intelligence
database cpus (cne interactive and six batch) and seven
terminal access systems (four TASs as described in theprevious chapter, three developed under IDHSC). The last
host is the ?etwcrk Ccntrol Computer.
B. BHSIMBE EBNTIROIU8UT
At the lowest level, the subnet capacity is 56 KB/s.
This is the physical maximum data transmission rate of the
telepbhcne lines. The boxes contributing to the effective
throughput are the communications crypto equipment, the
IaPs, the frcnt-end processors and the hosts. Each of thesecomponents have related software to make them perform their
network functions, and in the case of the cpus their user
services functions.
19
During the installation testing phase of the pack-t-
switching subnet in 1972, effective data throughput
measurements were in the 28-32 KB/s range. The configura-
tion for this testing included PDP-11 cpus connected to the
IMPs. Ill the PDrs had an LH/DH interface unit that
permitted connection to their IMPs according to the BBN 1822
specificaticns for interconnection of a host to IMP. Each
of the cpus had a simple data generation program. The range
of the thrcughput measurements was attributed primarily to
the differing cpu capacities of the PDP 11/40s, PDP 11/45,
and PCP 11/70. 7he next testing level had the same physical
configuraticn but different software environments. The PDP
11/40's were running under ELF-I. the PDP 11/45 was running
under BSX-11 and the PDP 11/70 was running under UNIX. For
each cf these systems, there was an Network Control Prcgram
(NCP) tc handle the host-to-host protocol as specified by
the IEPANET. There was also an application program that
served as a data generatcr. the effective throughput
measurements from this test were in the range of 15 to 20
KB/s. All of these throughputs excludes the host-to-host
protocol overhead.
Since these ueasurements were taken, two server-TASs and
cne lAS were added tc COINS. The TASs are configured to
handle 16 to 64 terminals. These are Teletype Model-40
terminals with CRT, keyboard and printer, and operate in
full-duplex at 2400 baud. The printer is slaved to the CET.
The one data-receive line is directed to the CRT.
C. SOPTUA22 INTIRONNINT
The software environment may be viewed as a trinity
consisting cf ths operating system, the applicaticn software
and the network software. At system generation time, t hese
three ccmponents define the number of ports with which the
20
cpu will perform netvcrk business. For example, a certainamount of memory is allocated for the operating system, theapplication software and the networking software. Insuppcrt cf natvcrk servic9s, a specific amount cf systembuffers is allocated. The number of system buffers in turndefines the number of simultaneous network connections a cpucan handle. Currently all TASs, including the server-TASs
have an interactive network capacity of 24 ports. SCLIS,the interactive datatase resource in COINS, has a netwcrk
capacity cf 15 pcrts.
21
° °, * o" ° ° '.' '. ........° ' , i , ° ." °, o".2, , .. °' ..& . . .
SCLIS, VSH and TRF offer interactive access. The
remaining six intelligence database hosts provide only batch
access. Both batch and interactive are discussed in thischapter.
A. BATCH SURVICES
Use of batch query systems involves a user inputting the
query at the local host and sending it to the remote system.
He then receives a job number or receipt for the query.
Some time later, which can range from minutes to days, the
query response is delivered to the terminal. Responses arepresented to TAS users only if they are logged on to a TAS
and specifically request to see the response.
B. INTERICTIVE SERVICES
The three funticnal interactive services available in
COINS-II today are: intelligence database (SOLIS), user-
support databases (USIS) , and managerial and admrnistrative
databases (BUIS) as explained below.
1. ~Databases
a. General
SOLIS is a partially formatted full-text search
and retrieval system. It contains the last 13 months of
messages and reports produced by the intelligence crganiza-tion. Searching can be done on the formatted fields ard on
the full text in any combination with the normal boclean
operators.
22
- . * . . . . * . .
k. Exchange Discipline
The data exchange between SOLIS and the user is
fall screen. After successful logon, the user is presented
with a form-screen where he can fill in the blanks with his
search terms fo: the full text and the formatted fields.
The user has three different form-screens to choose f:cm.
They are the aND-screen, the 0-screen, and the FREE-screen.
In the AND-screen, the search terms are ANDed together;
while in the OR-screen, the search terms are ORed together.
In the FREE-screen, the user may compose his own boclean
logic of the search terms. Within the AND- and OR-screen,
use of parenthesis is permitted to achieve combinations of
ANDs and CBs. after he has completed filling-in-the-blanks,
the query screen is then forwarded to SOLIS. The response
from SOLIS is again a full screen with the query the user
sent plus the number of messages that satisfied the search.
There is a blank area for the user to fill in, telling SOLIS
what should be done next. At this stage, the user can:
(1) refine or modify the query and send that again;
(2) request display of titles only;
(3) request display of the formatted fields of the
messages;
(4) request display of full-text of the messages;
(5) reguest a new form-screen;
(6) terminate the session; or(7) request forwarding of titles cnly or titles and
fcrmatted fields or complete messages.
2..L . ratabases
a. General
The User Suppcrt Infozmation System (USIS)
offers a variety of help functions to the network user. it
is an attempt at Computer Aided Instruction (CAI) but not in
23
-J
,.- . = - . " o - " .° * . o . . " •- . . - .. . . . . .. . .. , . . . . .. . -
the sase manner as the University of Illinois's PLATO
system. While PLATC interacts with the user posing ques-
tions and checking the user's answers, USIS is primarily a
one-way ccmmunication to the user. In USIS, the user can
ask for cne of several tutorials on the various languages
and data files in the network and learn how to employ TAS
commands and functions. While in USIS, the user can send
messages to and receive messages frCm User-Cantral
concerning aid he canit find in USIS, problems enccunter d
and ccmplaints. There is a limited amount of browsing and
no refinement of output capabilities.
b. Exchange Discipline
USIS initially sends the user a menu of the
available USIS-commands and waits for a response. The
exchange rule from the user to USIS is one character at atime. First-level ccmmands are norally terminated with a
NEWLINE. Lower-level commands are terminated with a special
sequence cf period (. and NEWLINE .on a line by itselIf.
Having reached this point, exchange between USIS and the
user is as fcllows:
(1) USIS sends a screen cf ata;
(2) user may respcnd with browse command of fcrard,
backward or return to the previous level;
(3) request for hard-copy must be done before reaching
this level. The user must indicate it on his first
ccm2and line.
The tutorials are basically copies of the hard-copy language manuals and file guides. They contain sample
queries and outputs, including badly constructed queries andtheir resultant error messages. It also contains a file of
latest-happenings in the network of interest to the user.
An on-line newsletter is not available yet.
214
- °- ----.
* -*-*3*,-~* nm u iz irr.-traiv Databases
a. General
The Netvczk Management Information System (!IIIS)
contains statistics cn file usages, numbox of queries,
number of atorts, size of responses and network problems.
1t prcvidres some basic matrix and chart displays. There is
limited browsing and no output refinement capabilities.
b. Exchange tiscipline
The lata exchange between NMIS and zhe user is
character at a time, which is terminated by the NEWLII4E ke-y.
After successful logcn, NNIS presents a menu to the use: and
asks for a response. After menu selection, NMIS prompts the
user fez search and display (viewed interactively orforwarded) criteria. This kind of exchange continues unt-il
HIS bas encugh information to proceed wi&th the act-ual work.
25
This thesis is that reconfiguring the COINS interactive
capacity intc one ccnsisting of collaborative interactive
and batch functions can be, under certain conditions, supe-
rior to the current form. We plan to describe the nature of
customer activity with a database as a combination of i6nter-
active and batch. It would make sense for the network to
accommodate this two-phased activity in a manner that offers
the best performance from both user and managerial perspec-
tives. The examination will be confined to the work prcfile
cf the .intelligence analyst. This has been prompted by the
abundance of empirical data in this area and the very little
information available for USIS and N3IS activities. The
procedure outlined bere can also be applied when invrsti-
gating user support and managerial activities. we will then
present a particular reconfigured network capacity scheme
whose performance will be ccmpared with the current method.
The available empirical data regarding TAS-custcmers'
usage of SCLIS was gathered from COINS. The data was
analyzed and statistical tests were performed to determine
the underlying distributions of arrival rates and service
times. The work-profile of a 'typical' network SOLIS
customer was derived from SOLIS logs. TAS logs Frovided
data on the percentage of user requests which required SOLIS
access. Appendix A ccntains the results of this analysis.
We implemented a computer simulation to aid in the
comparative analysis. The distributions developed from the
data were used to drive the stochastic models of the varicus
system ccnfigurations. Sensitivity analysis will be
performed with respect to arrival rates and transaction
service timzs. Th.se two parameters were chosen because of
26
the expectation by the COINS/PRO of increased customer popu-
lation and an in-agency study done on the SOLIS print4ng
requirements. This study's conclusion was that there will
he continued growth in demands for hard-copy output
(Ref. 6].
Measurements will be made of
1. average system time which includes service time and
wait time;
2. expected customer loss; and
3. proportion of interactive work.
The first twc measures are directly related to customer
satisfaction while the last may be of more interest tc the
network and database managers. Users are interested in the
amount of time to accomplish a job. This is the total
system time which includes both the service and wait time.
They alsc anticipate an available server when they arrive
for service. If the facility is busy when the customer
arrives, he is lost to the system. If the facility is busy
too cften when the customer arrives, it will discourage
system use and cause severe customer dissatisfacticn.
Althcugh tanagers are ultimately interested in customer
satisfaction, they also focus their attention on utilization
issues. we prcpose a proportion measure with respect to
interactive work. When evaluating an interactive queryrescurce, it is impc.tant to scrutinize and ascertain how
much of this facility is being used for interactivesearching. Putting it another way, we would like tc know
how much cf this facility is being used for non-interactive
processing, that is, batch work. Hence the prcportion
measure will give the percentage of session time which is
used for the search and refinement process. The coiplment
of this is the proportion of time used fcr batch work.
27
K .° .% - .. ..
VI. USER ROFILE &NALYSIS
A. GIUEBIL
Cur model of network structure involves cpu to cpu
connectivity as oppcsed to terminal to cpu connectivity.
This thesis is an examination of a network form which
involves cpu to cpu ccnnectivity for interactive information
sharing. We feel this is an important difference because weare dealing with a terminal interface that is capable of
intelligence, capacity and speed far superior to that of adumb terwinal. The cld adage that a chain is as strong asits weakest link is the same as saying that two devices can
communicate as fast as the slower of the two. The same
hclds true for the other two attributes of capacity andirtelligence. With respect to resource content, the commer-
cial database systems are bibliographic and abstract in
nature. This thesis examines usage of a full-text retrieval
system. ligingtcn has suggested that searching full-text tflarge documents may have a somewhat different pattern fromthe biblicgraphic environment [Ref. 7]. This simply meanswe cannot take full advantage of the work already done for
the ccmmercial database system with respect to detail work-
profile analysis. We will refer to these reports merely to
give credence to cur perception of the dichotomy of
service-time.
User activity must be described in terms that will makeit meaningful to the problam statement, i.e. non-interactive
use of an interactive resource. Customer service-time may
be classified into one of two modes. The first is interac-
tive, and is defined as a continuing dialog between man and
machine. when a user enters a request, he must wait to see
28
q
the system response before proceeding to the next request.
The othe: is called non-interactive and is defined as
follows: when a user enters a request for servic., he ne-d
not wait tc see the system response before he can prcceed to
the next request for service.
There is a plethora of studies on customer activities
with commercial and ncn-Defense interactive network infcrma-
tion rescurces and cnly a very few relating to full-text
retrieval databases. We will refer to these commercial and
non-Defense repcrts to show that we are correct in cur
percepticn of the user work profile. These investigations
ithus far have concentrated primarily on the interactive
nature cf bibliographic searches (Ref. 8], [Ref. 9],
[Ref. 10). [Ref. 11]. Their focus has been on search stra-
tegies, evaluating the impact of user training and
investigating methods to address multiple logon protcccls
and retrieval language problems. However, based on these
reportsr user activity can be viewed from a different
perspective. That is, while the user is connected to this
interactive informaticn resource, his activty may be cate-
gorized as either Intera ctive or batch. Included as
interactive are those user functions for search, refinements
cf search statements, perusal of hits, and if the databasemanagement system permits, narrowing down the hit-list
during perusal. Some studies have described this kind ofactivity as cycles within cycles [Ref. 7] cr a series of
sifting (Bef. 12] through the body of retrieved data until
the user is fairly satisfied with his 'find'. As stated
earlier, there have been numerous investigations into this
aspect cf how the user spends his productive search time atthe network resource. For completeness purposes, in
describing the user's total connect time, the studiesmention rsers commands for hard-copy output. In the commer-
cial database systems, where the charge is primarily
29
according to connect-time, there is a relatively small
amount of time devoted to on-line printout. However, there
is an indication that there may be a fair amount of cff-line
printing done at the database site (for a fee), and thatoutput is sailed to the customer. We define batch work to
te request for hard-ccpy output to the user site. As early
as 1973, informaticn scientists had been calling for an p
interface between batch and interactive in several areas,
one of which is the transmission of large amcunts ofretrieved data to be Eassed later via batch (Ref. 13].
All network access to the commercial information
rescurces are at dial-up speeds ranging from 300 to 2400baud. All custcmer access, with a few exceptions hhich we
will describe shcrtly, are through a variety of dumb dial-up
terminals. The exceptions are found in non-Defense govern-
ment agencies and at research institutions. The impetus of
these efforts was to address the front-end problems of
accessing a variety of retrieval systems each with their cwn
logon prctocols, query languages, and search strategies.
The wcrk in this area has been the development of an int'.l-ligent user interface [Ref. 14], [Ref. 15], [Ref. 16].Their aim has been to aid the user in formulating search
statements, and in scme cases, because of its knowledge of
the kinds cf information in the network, the inteiigent
interface will attempt connections at all hosts with the
database of interest until a connection is established.
Early work in this area was the development of COnnectcr for
Netwcrked Irformaticn Transfer (CONIT) by Marcus at H.I.T.
(Ref. 17]. Subsequent research using software based onCONIT has been carried on in Meadow's Individualized
Instruction for Data Access (IIDA) [Ref. 15], [Ref. 18].
The Network Access Machine (NAM) was developed at the U.3.National Bureau cf Standards (NBS) and Chemical Substances
Inforuaticn Network (CSIN) is now in place supporting the
activities cf the Environmental Protection Agency (EPA).
30
Although the study done by Dlmperic on SOLIS cOncen-
trates cn user query habits, there is still divisicn of
customer work into interactive and batch. Only 'C-e f
senticn is made in the study to hard-copy requests.
Although SOLIS has a large number of directly connected
user terminals, this thesis is interested in looking at the
netwcrk access and utilization of SOLIS. This is a database
of reports, with average report size of approximately 2000
characters. Several years ago, the American Chemical
Society (ACS) together with Bibliographic Retrieval
Services, Inc. (ERS), embarked on a series of experiments to
determine the usefulness of a full-text database (ACS
Journal articles) and their availaniliry for searching
online. The experiment was based on a relatively small test
database (Ref. 19], and looked only at the usefulness issue.
No mention is made of output demands. However, we de rct
feel this in any way invalidates our perception cf customer
activity with a database. When ACS and BRS move tc subse-
quent phases of the study with large test databases, we will
probably then see their reports referring to prin-, and
display ccmmands.
E. BCDEL DESCRIPTION
Tc provide a general framework for describing th .
cirrent approach and the proposed alternate, some elementary
concepts from queueinq theory are used. SOLIS is a facility
with 15 servers. Each server provides the same service at
identical rates. Custcmers arrive at the service facility
from the network at a certain rate. If customer arrival
follows a Poisson process and the service time is expcnen-
tial with parameter mu, this defines an M/M/15 queueing
system. In this simple example, all customers are ccllictad
into cne source. CCINS has four sources, one from each cf
31
*- - - - .• . .- . . --.- --.- -
the TASs. Figure 6.1 and Figure 6.2 represent the transi-
tion frcs cne populaticn pocl to four separate pools.
Statistical analysis and tests for the customer arrival
process tc the TASs, found in Appendix A, show custcmer3
arrive at the lkSs according to a Poisson process with
parameter lambda (i), ± = 1, 2, 3, 4. Arrival rates per hcur
)) S 0 L I S
ETICBKi
)cisson process 3 mu:
. ... servicelambda: arrival . rate
CUSTCHER rate
POEULATICS f-- ---- - - -
Figure 6.1 Cne population source to SOLIS.
at each cf the TASs are 17.68, 9.19, 5.10 and 2.53. TASs
are service facilities, offering network access, through
their 24 forts. Once at a TAS, only a certain percentage of
custcmers request service of SOLIS. The percentage cf users
requesting EOLIS access at each TAS is 0.70, 0.31, 0.098,
and 0.89. A customer request is granted only if there is a
44
32
: 1 - ;> TASpercentage to SOLIS17.6e TAL S0.70
c st ./ I -- - - - - -14
LI
--->1 cs 29.19 0.7, - - -
c ust./ ----- 1hour -- SOLIS...... I-- interactive
-- > I2
9.1 0.318
cust./hcur
. ..
us 2.631
r r
cust-/Ibcur 2 II
Pigure 6.2 Put population sources to SOLIS.
33
t .
free netwczk server cn his local TAS and a fee server at
SOLIS. The user at SCLIS may be engaged in only interactivework during a sessicn or may do both interactive and hatch
work in the session. SOLIS can nov be described as a
2-stage facility, where all customers enter stage 1 and work
there for a certin amount of time. They then prcceed to
stage 2 with a certain probability p or leave the system
with probability 1-p. Figure 6.3 illustrates this concept.
New customers can enter the server only if both stages are
empty. Fzcm the SOLIS monitoring logs, 0.24 of the
customers do only interactive work and 0.76 do both. Thq
distribution of work-time for only interactive work is expo-
nential with an average service time = 10.13 minutes.
Similarly, the work-time distribution for both interactiveand batch is also expcnential with an average service time =
12.92 minutes. For the purposes of the analysis, the seccnd
work-time was separated into the individual times for inzer-
active and batch. Statistical tests on the data showed thatthese distributions are also exponential with average
service time for interactive work = 6.2 minutes and average
service time for sending retrieved data = 6.8 minutes. Thisanalysis may be found in Appendix A.
C. TIME-LINE INSPECTION
With the definitions of customer service-time from
above, we now suggest a time-diagram which permits a graph-
ical view of the partitions of work-time. Figure 6.4 is one
such diagram. The begin and end times of a customer sessionare indicated. This session time is subdivided into inter-active and batch parts. Considering the amount of time the
interactive resource is occupied servicing this custcmer,then the proportion of interactive use is at the 0.6 level
or 60% of the time. Another very simple situation is shcwn
34)=3
* 4 4 -
4 ,
i exit
iiSOLIS Sevice Facility
.* -. *4.- ... .. - .. ... -. . - -
st e 2 1-P
(k aNeh)
,I,
~(or)
st I II1
+--act-i
I
Pigure 6.3 2-Stage service facility.
35
"--------
'pI
<7]
XzRXxIzxxxxxxxxxxxx lxxxx I
0 510begin end1
I
in tezract% :ive X XXXXX
batch
I
Figure 6.4 Tine-Line diagram of I retrieval session.
in Figure 6.5 . The three horizontal time lines reflect the
facility's ability tc handle up to three customers at a
time. Time again is divided into its respective work modes
of interactive and batch. This case is simplified by having
all three users starting and ending at the same time. The
propcrticn cf interactive use is 0.40. The number of
custcmers the system was able to service in these ten time
units is three and their average service :ime is ten unit.
The real world dces not operate in this manner. 7hese
diagrams were used tc give some insight into the problem at
hand. The ccmputer simulation has incorporated the vulti;l
TASs, their customer arrival dist. ibuaions, percentages of
SOLIS requests and the SOLIS work profile distributicns.
36
. i .. 'o - -o .. - . . Q o 1 , m° , . , . . . . q . ,o ' j - .. - - - .. •
xxxxxxxxxxxixxxxxxxxxx
--- * -. *
0 5 10
interactive: XXXXX
batch -
- -- ,.. - --- .. ___'_ __ _ _ _ __ '_ __ __ __-' I
Figure 6.5 Time-line of 3 retrieval sessions.
37
MILO
vII. hi &j H.I!I gg .qJ€: o !i o I!I .RcTrRAN
A majcr drawback of the current method is that SOLIS is
set up as a 2-stage facility. Stage 2 is purely batch in
nature and the server is kept busy transmitting data at a
comparatively slow s;eed to what the server believes to be a
terminal. While the customer is in stage 2, no new customer
can enter the server to begin his stage 1 processing. With
a substantial population of users, most of whom request a
fair amount of data to be transmitted back to their TAS,
either for printing at their local terminals or for further
manipulation at their local TAS, we can easily foresee scme
problems. One way cf addressing this kind of situaticr is
to provide a high-speed background data-transmiasion
facility between the server-cpu and user-cpu. Peferring
back to our model of SOLIS, we propose a transformaticn from
a single ncde, 2-stage facility, to a 2-node tandem netwcrk
as shcwn in Figure 7.1 Each box in that figure describes a
queueing system consisting of a queue and serve-(s). Within
each box is given tke node number. Node 1 represents the
interactive facility with 14 servers and node 2 is the hatch
facility with 1-server. The original 15 interactive ports
on SOLIS are re-allccated to 14 interactive and one batch.
When a customer has completed work in node 1 and has gener-
ated data to be sent back to him, his work reques ia
forwarded tc node 2 for processing. This arrangement now
leaves node 1 free for interactive work.
The price for this design is that we usurp one cf the 2(4
servers on TAS and cne of the 15 servers on SOLIS. Using
cur simplified time-line diagram again, Figure 7.2 illus-
trates hcw three users can be accommodated on two
38
- .. . ,... . .. . . .. . . . . . . .
exit I I
node 2 itit, I
pr'cceed tonode 2
(or)
node 1I I
, I
11 2 3 14-----------
ent e r
figure 7.1 SCLIS as a 2-node tandem netvork.
39
.
interactive paths and all their data transfer requests on
II
IXXXXXXXXXZXXXxZXxxXXXXXXXXXYYYYYY!!YYYYYYYYY I
0 5 10 1ZZ222222ZZ Z
0 5 10
0 5 10begin and
I Iinteractive resource: --------
hatch resource:- -
interactive work: XXXXiYYYYY!zzzzz I
batch work:
Figure 7.2 Tine-line with 1 data and 2 interactive paths.
the one high-speed link with capacity to spare.
The ctber expense would occur in those cases when the
user wants the data printed at his terminal as soon as
possible. In the cutrrent situation, when the user gives the
command to print the data, the system then begins tc deliver
the data at his terminal-printer at an approximate rate of
2400 baud. In this new proposal, the user must wait fcr the
Cpu 6t cpu transfer (at an approximate rate of 20 KB/=) and
any queue time at node 2 before seeing any data at his
printer. This particular side-effect must be considered
40
-, , .. ., ., . . - . .
very carefully in tlke evaluation ,bacauseo this may bs too
much ef an inconvenience to the cUStOMer.
41
VIII. IBE Muon_ SER!ICOS HODE1
This chapter will discuss the model's goals, functional
design, internal design, preliminary results, "validation and
appl icaticns.
1. CESIGI GOALS AND CONSIDERATIONS
The Network Services Model is a discrete event simula-
tion that models netwcrk resource allocation in response to
arriving customers' requests. It does not attempt to model
the irternal operaticns of the server hosts and user opus,
nor the particular flow of messages and packets through the
COINS-II subnet. Instead, it focuses on modeling the
network from the point-of-view of how the natwork capacity
of the server and the user Cpus is consumed in support of
interactive database query and retrieval services. It si~u-
lates customer arrivals at the user cpus, their request for
interactive network access, the allocation of network
resources if available for the session, and their
de-allocatien at session completion. Design and implementa-
tion ware motivated by the following seven design goals.
1. The mcdel should be a faithful representation of the
network entities contributing to interactive services.
2. The model should serve as a realistic simulat.icn of
customer activity and allocation of resources based on
this activity.
3. It should be able to take the same customer activity
and allocate resources based on the alternate
proposal.4. I-s shculd provide metrics for performance compazison
between the twc approaches.
42
. . . . . . ... . . .
711
5. It shculd serve as a measurement tool fcr doing snsi -
tivity analysis as arrival rates and work prcfile
distributions change.
6. As we acquire further insight into custcmer work
habits (substantiated by more detailed logging and
mcnitoring data), the model should be flexible enough
tc acccamodate these statistics.
7. The model should be extensible and easy to mcdify so
that it can serve as a long term design tool for the
COINS/PHO.
Cne cf the most important objectives of the model is
that it he a faithful portrayal of the allocation of netwcrk
capacity by the TkSs, server-TASs and HOST when providing
interactive network services. It is essential that the
model behaves in a manner consistent with the flow of inter-
active Jcb requests through COINS-II. Interactive jobrequests are initiated from a TAS or server-TAS, never from
the HCST. For example, customers arrive at one of the TASs
or server-IASs according to some distribution. Having
arrived at a TAS, the customer then requests services tc cne
cf the interactive resource in the network. Customers at
any TAS may request services at any of the server-TASs or
HOST. Cn the other hand, customers at any server-TAS may
request services at the HOST or any of the server-TASs
except its own. If there is an available port at the
user-cpu and the database-cpu, the demand is honored and the
apprcpriate resources are allocated for the duration of thesessicn. These events should occur in the model in the same
fashion that they do in real life.
Statistics ccncezning customer arrival and their ngtwcrk
requests ware gathered from the accounting logs of the
user-cpus. The model, therefore uses these distributions
for the generation of network events. Characteristics of
the sessicn are anotker important aspect of this study. The
43
empirical data collected from COINS-II offers statisticallysufficient informaticn only for the SOLIS database host.Hence our particular analysis will be confined to evaluating
network utilization Cf one database resource. It must be
pointed cut that this is not a limitation of the simulation
model. It will support up to n-database resources, with thesize ¢f n dependent only on the size and capabilities of thecomputer the simulaticn model is run on. The descripticn cfthis simulation will ke of its full capabilites.
with the main issue being the comparison of two methods,
there are two approaches to model implementation. One wouldbe tc implement two simulations, each reflecting a parti-
cular strategy; cr implement one simulation usingappropriate flag setting to regulate the simulation control
flow for one strategy or other. Since only one aspect ofinteractive processing is changed, the later method is used.It was felt this is tetter than having to contend with main-
taining two separate programs.Since the model's main purpose is tc furnish perfcrmance
measurements of the two approaches, it must be able to take
the same set of distributions and work profiles and execute
for the current system and then execute for the alternate
strategy. These distributions and work profiles of thedatabase-cpus are input parameters to the simulation,
thereby giving the model some level of flexibility.Performance measurements are done in the two general areas
cA number of customers refused and average system time. Themodel can be run with the provision of queues for custcmersawaiting network access. When the model is run with queues,further measurements are taken for the average wait time andthe average wait time, given the queue is not empty. With
respect to the alternative proposal, average system time ismeasured cn two levels. One measure incorporates the time
it takes tc get all the output printed at the customer
.dl 44
terminal, (sorvice-timel) while the other measure inccrpo-
rates the time it takes to get all the output only tc the
user-cpu (service-time2). We feel it is important to make
this kind of distinction because of the variety of int-nded
W ujes cf the retrieved data once it arrives at the user-cpu.
The simplest activity is the mere printing of the data atthe user-site. However, as reviewed in the literature and
in light of the en-going work by the COINS/PMO, there is a
definite shift from straight printing to some data massaging
and scme early efforts in data fusion. There arq no machine
logs available to indicate to what degree this is occurring,
so these twc statistics are computed to provide the range of
possible expected system times. System-time2 is important
also because it provides an indication of how much sooner a
network path becomes free for re-use.
To be a practical design tool, the model should be able
to be used by the COINS/PMO and its personnel to investigate
the impact on customer services as the network grows with
respect tc more user-cpus and more interactive database-
cpus. The analysis in this thesis is based on the current
configuraticn and wcrkload in COuL-II. However, in ths
next 12-mcnth period, COINS expects to introduce two acre
TASs into the netwprk and two more in the next twc years.
The model should be able to accommodate such changes in
network configuration and workload. For this reascr, infor-
mation ccncerning each database-cpu and user-cpu ar.
specified as run-time parameters to the model. The number
cf HCSTs, TASs and server-TASs and their respective
profiles, including their network capacity and customer
arrival rates are part of the data read in atc run-time.
This flexibility to adapt to network changes was a major
influence for modular implementation.
'45
SINSCEIET 11.5 was used because it is equipped %ith the
mecharics for handling discrete-event simulations and hat
much of the versatility of a general programming language.
It has the traditional concepts of permanent and temporary
entities, ownership and membership in sets and queues. To
add a new TAS to the model, a minimum of three changes must
be made to the simulation. A new ARRIVAL-event must be
added tc the structure, a command to initiate the start of
arrivals for this new TAS, and finally, at the close cf the
simulation, there most be a command to terminate the arri-
vals for this new TAS. If the interarrival distributicn of
the custcmers to this new TAS is the same as one of the
existing 1ASs, but with different arrival rate, then thesame call can be made, using the different parameter. If
however, the distribution is different, it is only a matter
of writing a routine describing the distribution and calling
this new routine for the next arrival. For the database-
cpus, the cnly new work that would be required is when itswork-profile distrihution is lifferent from any of the
cthers already in the model. If i is different, a similar
procedure must be followed as was described above fcr a new
LAS.
B. HIRECRBE AND SOF2UARE ENVIBCNKENT
The Interactive Services Model runs on the IEM 3033
Attached Erccessor System under OS/VS2 at the W.R. Church
Computer Center, Naval Postgraduate School. The software is
a SIMSCRIPT 11.5 program which has approximately 1,350
executable source language statements. Work areas are
dynamically created during execution, depending on the input
parameters. At-ention must be given to the Job Control
Language (JCL) setup with respect to execution time and
storage requirements for the job. Runs for -his thesis were
46
I
defined as a CLASS G job, permitting 15 minutes cpu time.
The source code is the property of the U.S. governwent.
Anyone interested in possible use of the program should
contact the author. Operating instructions for the prcgzam
can be fcund in Appendix B to this thesis.
C. PONCTIONIL DESIGN
The ucdel specifies when certain events are to occur,
based on the distributions given for each event. Ir. the
alternate method, the queueing discipline of the high-speed
facility contributes in deciding when departure events are
to take place. The SIBSCRIPT 11.5 timing routine ac-ually
handles the clock advances and the firing off of event
processing according to schedule.
After initializing its internal tables based oninput parameters, a customer arrival is scheduled for eachcf the user-cpus, and the simulation begins. There are a
total of 13 events that can take place in a simulaticn run.Five events handle the arrival of customers to each of the
user-cpus, and one terminates the simulation. The start
state of the model is defined as no customers and no inter-active network capacity is being used. Parameter input
specifies the available resources for interactive servicesfor each of the cpus. The remaining seven evets are
concerned with the sequence of events that begins with the
cus-cmer requesting an interactive network service tc theissuance of a command to initiate the retrieved datatransfer and the steps involved in execu-ting that command tothe user-cpu, then to the database-cpu and the subsequent
transmissicn of the data back through the network to theuser terminal. The section on event handling prcvides a
47
8* " '"" - _ i ; -,. ; ,.".".",•-". -".". . ". •" .. " . . . . .. ,".. ."-"., ."., " ".• •.
e. i n i'U4*i7.. . . . . . .. T *
4T-
more detailed description of these internally genera-adevents. Each of the user-cpus have their own arrival ratss
whose inter-arrival times follows some distribution. Thekind cf distribution and its parameters can be specified as
input parameters. Analysis of the inter-arrival times from
the empirical data shows that its distribution is expcnen-tial. The SINSCRIPI 11.5 statistical distribution packagesoffers a fairly wide range of distribution functions to
choose fzcm. They include erlang, gamma and beta to mentiona few. For a comprehensive list the reader shculd see
[Ref. 20].
2. IL __r. ge.ction
Selection of which interactive database cpu is also
derived from the empirical data. Throughout the simulaticn,
the mcdel maintains the currently available nqtwork capacityfor each component. If there is sufficient network capacityat the user-cpu and the requested database-cpu, the appro-
priate resources will be busy for the duration of thesessicn. If there is insufficient or no facility free, the
request will either be refused or placed in a queue,
depending cn the run-time parameters. Each time resources
become free, the earliest job request in a queue matching
the available capacity is selected for processing.
3..2.; _ ce Pr o f 1
Icg data from each of the database cpus were used to
define the session Frofiles of interactive service times,data transfer times and the percentage of custcmers
requesting data transfers. These parameters can be modifiedat run time without changing the program code. Two pcssible
things can ccur at this point. if the customer has onlyinteractive work to perform, the resources are tied up for
just this period of time. However, should there be demand
48
m m m mmIm m m
lmlm m m( - m m l- /mh = ---- , .. . . .
for print-data, then the network resources are kept busy fcr
the duraticn of the interactive portion plus the data
mransfer portion. The time defined for the data t:-ansfer
portion is based on the distribution of output character
size, transformed tc number of bits divided by speed cf the
terminal. For example, if the number of characters is
50,000 and each character is transmitted as an 11-tit code
with the terminal speed as 2400 baud, then the time for the
transfer tc take place is
(50,OCO x 11) / 2400 -229 seconds = 3.8 minutes
Pcr the alternate proposal, when print-data is
demanded, the interactive resource is freed for further
interactive work, and the data is sent en the high-speed
facility if it is free. If the high-speed facility is notfree, the transfer request is placed in a queue until such
time as the resource becomes available. For the purpcs.s of
this model, the network potential is estimated zc be 20
KB/s. In the example given above, the transfer would take
(50,OCO x 11) / 20,000 = 27.5 seconds = .45 minutesThis specific implementation permi-s three different
distributicns which are uniform, normal and exponential.
This was done to indicate to potential users of the mcdel,
that the mcdel is rct restricted to only the expcnential
distribution, and that adding a new distribution is a simple
exercise because the program only calls the statistical
distribution functicns of SISCRIPT.
Although this thesis is primarily concerned with how
to get more interactive work done on the interactive
resource, it must still consider how long it will take forthe customer to eventually get his product. In the scenario
just described, the data will be at the user-cpu in 27.8
seconds; however, it is not at the user terminal. And
furthermore, that 27.5 seconds is straight transit time and
does not include any queue time if the transfer request had
49
% , % . . - . . • •° .
to wait it a queue. Should the customer want the data at
his terminal, it would take another 3.8 minutes for it to betransferred from the user-cpu to the terminal. It is £umpor-tant that the model take these issues into consideration bykeeping statistics on these different system times sc that afairer ccparison can be made.
4. f._-1pled gueueig pjgscipline
In the real system, when the network resources arebusy, customer requests for SOLIS access do not wait in thequeue; they are lost to the system. For the alternatemethcd, queues can develop at the high-speed server. Its
queueing discipline .s described next.Arrivals to this one-server facility can come from
four different population sources or TASs. When the task
arrives and the server is free, the data is transmitted
immediately to the appropriate TAS. However, if the serveris busy when the task arrives, the task is placed in a queuecf work des~ined for the TAS from which the work originated.
Once tk9 server has started transmitting data to a TAS, itwill continue to so until the queue for that TAS has been
emptied. For example, let there be three TASs, denoted byTASI, TAS2 and TAS3. Furthermore, let there be two data
files for each of the TASs that the high-speed facility must
transfer. The server will begin work on the TS-queue whose
task arrived the earliest. In this example, let the task in
TASI-queue have the earliest arrival time. Then the server
will begin data transmissicn to TASi first. When that is*: completed, it will proceed to the second task in the queue
for TASi. If a new task arrives for transmission to TASI
before tte server has completed servicing the first twotransmissions, the server will proceed to work on task threeafter it has completed the first two transmissions. Say nowthat the server has completed task three and no new work has
50
Ft.-1
arrived fcr transmission to TASI, then the server will pick
the next earliest task waiting in the TAS2-queue and the
TAS3-gueue.
D. INTERNAL DESIGN
1. Cnzerall §_rcu
Using SINSCRIPT 11.5 has given the INS model a very
simple control structure. The permanent and temporary qnti-
ties and their relationships to each other are defined in
the PRIEAELE section of the model. All the events, their
attritutes and priority handling are also declared in this
secticn. The global variables and any counting and aver-
aging are specified here. Program MAIN is concerned only
with managing the general flow of control including the
setting up of the initial system state and providing the
starting events that will set the simulation in motion. The
simulation is not begun until an explicit statement START
SIMULATIC is issued by MAIN. At that point, control is
transferred to the SIMSCRIPT 11.5 timing mechanism. The
timing mechanism manages the flow of control from event toevent as they are scheduled to occur. When an eventprocessing has completed, the timer searches through theevents-list, locking for the earliest next-event to sche-dule. The timer then updates the system clock and transfers
control tc the event routine. When no further events arefound on the events-list, the timer returns control back to
MAIN.
To illustrate this point, the following example and
rccompanying seven Figures are provided. Let there be two
TASs, denoted as TASI and TAS2. Suppose TASI has two
network pcrts and TAS2 has three network ports, and let
SOLIS have two netwcrk ports. Figure 8. 1 is a picture at
the start of the simulation, the clock is at time - 0
,5
4. 5 1
".
... ........-...........-...... -, . - . . ... - • .- . .. . .
minutes. All the resources are free and are indicat.d by
empty boxes. In Figure 8.2, the clock has advanced to time
* 2 minutes and shows a customer has arrived and requested
network access to SCIS. The Figure shows a path estab-
lished betwsen TkSi and SOLIS and the allocation of the
network Ecrts. The Fort boxes are Xed. Customerl is in an
interactive session with SOLIS. From customer work profile,
it is determined that customerl will have an iteractivesessicn cf 6 minutes, followed by request for hard-copy that
will last 7 minutes. &ccording to the interarrival distri-buticn of customers, customer2 arrives at TAS2 at time = 5
minutes, with a work profile of 6 minutes of interactivework and no hard-copy request. Figure 8.3 shows the state
cf the mcdel at time = 5 minutes. There are now two paths
to SOLIS, one from TASI that started at time = 2 minutes andthe seccnd one that started in this time. The next event to
occur is at time = 8 minutes when customeri at TASI
completes his interactive work and now goes intc batch
processing. This is indicated in Figure 8.4 and the path
doing batch processing is indicated as B. Another customer
arrives at TIS1 is the next event that occurs at time = 10
minutes. Figure 8.5 shows an attempt to establish a path
tetween TASI and SOLIS, but is not successful because of
insufficient resource at SOLIS. Customer3 is lost to thesystem. The next event occurs at time = 11 minutes when
custcmer2 has completed his interactive work with SCLIS.
Figure 8.6 shows the path between TAS2 and SOLIS is now
free. The next event occurs at time = 14 minutes whencus"cmerl has completed his batch processing. Figure 8.7
now shows the path between TAS1 and SOLIS is now free.
52
. .. . . . . • . . . .. .
1Y ~t -e--- - I .
time 0
TIS1 SOLIS
I -I I I
TAS2---- ---
1-+ Event 1 at time 0
__ __ __ _ __ __ _
Figure 8.1 Initial State.
time 2
TASI SOLIS+-- custcmer 1lxi-- - - - ----- ---- +
TAS2 4 -----';"I I
Event 2 at time = 2.. : -- +-Int era ctivg4 session:"-, I Port occupied : XX
Figure 8.2 State 2.
53
-7 .
time =5
TAS1 SOLIS+--+custcmer 1 +--+-xx ----- xx.
I I /1/IX I+--+ I+--+-/
TAS2 /+--+custcmer 2 /
Event 3 at time = 5 minutesi iInteractive session:
Port occupied : XX
Figure 8.3 State 3.
time = 8 minutes
TA51 SOLIS+--+custcmer 1 +--+-tXXI EEEEEBBBBEBBBEEBBBBBEBBEBBBBB IXXI+--+ +---+-
I I I///IXXI-vent / time -
+--+custcmr 2 /I Xx III///
Event 4 at time = 8 minutesi i-- - I~nteractive session:I I :1/I///
Batch vork : BBB.BPort occupied : XX
Figure 8.4 State 4.
54
= |-
. . . .. . . . . . . . . . . . . . . . . .. . - •.. .
time - 10 minutesTASI SOLIS+--+customer 1 -+I XXI EEEBEBBBBEBBBEEEBBBBEBBBBBBBBB IXII
IXlcustcmer 3, blccked //1/11+-/ -- ,-
TAiS2/+--+customer 2 !+-- t Event 5 at time = 10 minutes
Interactive session:
Eatch work : BBBBBEcrt occupied : XX
Figure 8.5 State 5.
time 11 minutes
TAS1 SOLIS+ --+custcmer 1I XXIEBEEEBBBBBBBBEEBBBBBEBBEEBBBBBIXXII I I I
TAS2----
I I+--+. Event 6 at time = 11 minutes
+--- Interactive session:I I Eatch work : BBBBB
+- -__crt occupied : X
Figure 8.6 State 6.
55
time = 14 minutesTAS1 SOLIS---- +--- -I I I I
I I I I+----+ -- ,-
T&S2+----I I+.--, Event 7 at time = 14 minutes
Interactive session:IEatch work : BBBBB
Port occupied : XX
Figure 8.7 State 7.
For each set of network configuration, four runs are
made. The two major categories are the current and the
alternate proposal. And within each of these cases, the
model is run two times, one for the situation where
custcmers gc away when the netwcr.k resources are not avai-
lable; and the other, where they are placed in a queue. All
queues axe considered to be first-in-first-out (FIFC).
Sequencing through these four iterazions is managed by MAIN.
2. j_ f j 2 " ij .
There are 13 events that can occur in the mcdel.They will be described in chrcnological order. The reader
is referred to Appendix F of this thesis for the logic
diagrams cf these internally generated events.
V"
56
:,
a. Events 1-5: TASX.ARRIVAL, where x - 1, 2, ... ,
5
There are five events that handle the customer
arrivals at each of the user-cpus. The work perfcrmed inthese events are:
1) schedule the next arrival according to the distribu-
ticn for this user-cpu.
2) determine which interactive network host to request
services.
3) determine the wcrk profile at this interactive host.4) if the current network capacity permits, seize the
atrcpriate resources.5) if there is tc be no print-data command, schedule a
departure at tte end cf the interactive portion.6) if there is to be print-data comaand, schedule the
event to handle the priat-data command.
7) If the current network capacity is not able to satisfythe request, file the request into the queue for new
work cr ignore the request, depending on the inputarameter.
8) update the appropriate statistics gathering variables.
b. Event 6: THDEPART
The next event is the departure from the systemat the end of the interactive portion. The work perfcrmedhere are:
1) release the interactive network resources.2) If there are any other departures of this same nature
in the events-list that is to occur at this same timeinstant, process this event by releasing the interac-IL> tive network resources used by this event.
3) Having updated the network availability, searchthrough the queues of new work for any job request
57
that can be satisfied and schedule the appropriate
events. If there are no job requests waitirg in the
queue, return tc the main timing routine.
4) Update the appropriate statistics gathering variables.
c. Event 7: USEND
Event tc handle sending print-command to the
user-cpu. Schedules the event at the user-cpu to handle the
print-ccmmard in the amount of time to send the command fromthe custcmer terminal to he user-cpu.
d. Event 8: UC.ARRIVAL
The event to process the print-command at the
user-cpu performs the following:
1) If the user-cpu is busy handling another command, file
tle request in the queue for the user-cpu.
2) If the user-cpu is free, set the user-cpu flag to busy
and schedule tte event to release user-cpu rescurce in
the arcunt of time to process the request.
e. Event 9: UC.DEPART
The event to free the user-cpu after processing
the print-ccmand does the following work:
1) Set the user-cEu flag free.2) Schedule event to handle print-command at the
sezver-cpu.
3) If there is more work in the user-cpu queue, take the
next task, set the flag to busy and schedule the event
tc release user-cpu resource in the amount of time to
process the request.
4) If the queue is empty, return to the main timing
routine.
58
4%*t4 y* ~ .- * - . -
f. Event 10: SC.ARBIVhL
Event at the server-cpu to handle the prirt-
ccmmand dces the follcwing work:
1) If server-cpu is busy handling another print-commard,
file the request in the queue for the server-cpu.2) If the server-cpu is free, set the server-cpu flag to
busy and schedule the event to release the server-cpurescurca in the amount of time to prepare the date fortransmission.
g. Event 11: SC.DEPABT
Event to free the server-cpu after preparing the
*data fcr trarsmissicn performs the following work:1) set the server-cpu flag free.2) Schedule the event to send the data to the user.
Current method: Schedule release of in-neractive rescurce
in the amount cf time to transmit the data.
Alternate methcd: Schedule release of the interactive
resource now and send the requested data on the high-speed facility if it is available, otherwise place it
in the queue of work for the high-speed facility.3) If there are pending print-command tasks in the queue,
work cn the earliest task, set the flag busy and sche-
dule the event to release the server-cpu rescurce in
the azcunt of time to prepare the data for transmis-
sion.4) Fcr the alternate proposal, release the interactive
resource and icck for other events that are ccmplexed
at this same time.
59
4w
.--.
h. Event 12: LTHDE PANT
The event handles the departure from the system
of those who requested data transfers. The appropriate
statistics gathering variables are updated. Based on what
* departure has occurred, the appropriate network resources
are released. However, for the alternate configuration,where we are using a high-speed facility to pass all print-
data output, this particular resource is not released until
its queue is emptied.
i. Event 13: CLOSING
This event cancels the scheduled TAS arrivals.
SIMSCRIPT 11.5 provides a framework for handling
concepts in simulaticn such as permane-nt and temporary enti-
ties, queues and events.
a. Permanent Entities
There are four kinds of permanent entities. The
first twc are HOST and TAS. The server-TAS is included as
both a HOST and a 7AS because it really serves these two
functions. The important attributes for the HOST and TAS
are tkeir taximum number of network ports and a flag-fieldto denote when it is a server-AS. Since th. model handlesin detail, the sequence of events beginning with the user
issuing tte prin t-command, additional attributes of a busy-
flag and a queue have been defined.
To handle TAS to HOST connectivity, a permanent
entity called TASHOS7 is defined. The important attributes
of this entity are the maximum network paths between a given
TAS and a give.i HOST and identification of this given TAS
and giver HOST. For example, suppose there are two TASs,
60
. . . ,, . . ~. * .* . .. -. .. . . . . . . - - .. . + . . + . .+ . .+ . + .+ . +. . - . - .
.7
called TAS1 and TAS2 and one HOST. TASI has a capacity for
tan interactive ports, TAS2 has a capacity for 25 inte=ac-
tive ports, and HCST has a capacity for 15 interactive
ports. This results in two permanent antities called
TASHOST. The first cne is for connectivity between TAS1 and
HOST with a maximum possitle capacity of ten interactive
paths. The second TASHOST is for the connectivity tetween
TAS2 and HOST with a maximum possible capacity for 15 inter-
active paths.
The fourth type of permanent entity is same in
concept as TASHOST and is called LPATH, reflecting the
high-speed data transfer facility between a TAS and a HOST.
LPATH has a queue and the attributes to identify which TAS
and which HCST.
t. Tempcrary Entities
Temporary entities are created and destroyed
during the course of the simulation and are called tasks.
They are created only when a request for service cannct be
honored because the service facility is busy. They are
placed in a queue and remcved only when a server becomes
free. All queue disciplines are first-in-first-out (FIFC).
There is a potential for four different kinds of tempcary
entities that can be created during a simulated run. They
are:
TASK: Created when there is insufficient network capacity
tc support an interactive session. It is placed
in the appropriate TASHOST queue.
UTASK: Created when the TAS is busy handling ancther user
request to send the print data commanI to the
HOST. It is placed in the queue for the TAS.
STASK: Created when the HOST is busy handling ancther
user request to prepare data for transmissicn to
the iser. It is placed in the queue for the HOST.
61
"," ** .. . -' 4*" " " . . . : , - .".. - . . . , .i . . -, ' -. . .. ., - --.
LTASK: Created only in the model of the alternate
proposal. It is created when the high-speed
facility LPATH is busy servicing another transis-
sion request. It is placed in the appropriate
LPETH queue.
c. Parameters
To make the simulation as flexible as pcssible,
the program has the mechanisms for describing the dc-sired
network configuraticn and characteristics at run-time.HOST characteristics include:
1) number of interactive ports;
2) prcportion of customers doing only interactive work;
3) for customers dcing only interactive work, the distri-bution and its parameters which describe this servicetime; The simulation expects service time in minutes.
4) for requests of hard-copy output, the distribution and
its parameters describing the amount of characters
that Js to be transmitted; The simulation expects the
number of characters and will make the transfcrmationinto the amount of time to transmit the data.
5) for sessions wbere a user will do both interactive and
batch work, the distribution of the interactive
pcrticn of the session and its parameters describing
this service time. The simulation expects :nteractive
service time in minutes.TAS characteristics include:
1) arrival rate of customers to the TAS; The simulationexpects this to be in the number cf customers per
hour. Furthermcre, the simulation assumes this to bea Ecisson process.
2) number of interactive ports;
3) of the customers arriving, the proportion which willrequest some network access;
62
*.1
4) for those TAS customers requesting network access, theproportion of users using each of the network database
facilities;
,. PRBLININARY RESULIS
There is particular interest in the comparison of
performance as the arrival rates and the amount of data tobe transferred at the end of an interactive session is
increased. The methcdology adopted was to run the model
with the empirical data from COIS-II and establish that asthe baseline. This baseline consisted of four TASs and oneinteractive host. All four TSs have 24 available ports for
network access. Table I shows the parameters of the base-
TABLE I
TAS ojerating characteristics
arrival % toTAS 0 # ports rate SOLISTAS 1 24 17.68 0.70TAS 2 214 9.19 0.31TAS 3 24 5.10 3.098TAS 4 24 2.63 0.82
(c ustcm ers/hour)
line configuration of the four TASs. Customer arrival rates
to these lASs ranged from 2.63 to 17.68 customers per hour.The proportion of customers selecting to go to SOLIS rangedbetween 0.70 to 0.82. After establishing this baseline,several runs were made increasing only the arrival rates of
63
n nm n m mn nm m m -. ' " ****.- =' .h * - .-;' m '.. . . . .. .. " " . . . ," "."
TIAS2 to AS4 until they all reached 17.68 customes per
hour. The next series of runs started with the four TASS
and progressed up tc eight TASs. All arrival rates were
17.68 custcmers per hcur.
The cther variable of interest is the amount of data to
-. be transmitted on this interactive connection. The appcrach
to this was similar to what was done with the arrival rates.
Starting with the original empirical data, subsequent runsinvolved increases in the data transfer amount. The initial
run was for 89,018 characters. This was increased by 10%incresents tc 209,900 characters.
The next series cf runs involved the additicn of newTASs to the network first starting with the original data
*transfer demands and proceeding up to 209,900 characters.
In this third series of runs, each of the TASs had an
arrival rate of 17.68 customers per hour.
Table II shows the results of the baseline configuration
and wcrklcad as the arrival rates wers varied. They dc not
show any problem with the expected customer Icss. The times
A given are in minutes. In the alternate method, two kinds of
system time were measured. System timel includes the tran-
smissicn of data to the user terminal, while system time2
considers the service completed as soon as the data is
received at the TS. As expected, system timel is alwayslarger than the system time for the current method. The
differences range from 44 to 52 seconds. On the other hand,
comparing s~steu time2 with the system time of the currentmethod, system time2 is smaller by about 4.5 minutes. (range
of 4 minutes 30 seconds to 4 minutes and 42 seconds).
Graphing these results in Figurs 8.8 presents a better
picture of where systes tinel and system time2 lie in rela-tion to the system time of the current method. With respect
to user services, Figure 8.8 indicates that, if the customer
wants the data printed at his terminal site immediately, he
64.°9
N,N . . . . . . . .
I r . ,l Ill ''l '" I l " iii Iiii iijI
I .I
I IIi I
.I I - . I
I" .
I. I I" II ' l II - "- ,- I --± " . .. I -"
TABLE II
Increasing arrival rates
CHANGE CURSINT ALTERNATEIN SYSTEM EXPTD SYSTEM SYSTEM EXPTDARRIVAL TIME LOSS TINE1 TIAE2 LOSS
1 12.43 0.0 13.17 8.01 0.02 12.39 0.0 13.17 8.01 0.03 12.41 0.0 13.18 8.00 0.04 12.42 0.0 13.21 8.03 0.05 12.143 0.0 13.23 8.03 0.06 12.38 Q.0 13.19 8.01 0.0712.35 8:01 13.17 8.02 0.0I812.36 00 13.20 8.04 0.0912.35 0.004 13.21 8.05 0.01
ARRIVAL RATES1 17.68, .1, 9.190 2.63 CUSTOMERS PER HCUR2: 17.68, 17., 9.1 5 10 CUSTOMERS PER HOUR3 17.68, 17.68, 10.11, 5.61 CUSTOMERS PER HOUR4: 17.68, 17.68, 12.02, 6.20 CUSTOMERS PER HOUR5 17.68. 17.68, 14.80, 7.47 CUSTOMERS PER HOUR6: 17.68, 17.68, 17.68, 9.04 CUSTOMERS PER HOUR7: 17.68, 17.68, 17.68, 12.02 CUSTOMERS PER HOUR9: 17.68 76 17.6, 14.80 CUSTOERS PER HOR9: 1L 8 AS.6 17568 CUSTOSERS PER HOUR
times are in minutes
will have to wait abcut a minute longer in method 2 than he
does currently. On the other hand, if the customer is not
interested in printing the data immediately, or wants to
have it merged with cther query results at a later time, he
can be ccipleted with his work about 4.5 minutes sooner in
method 2 than he does currently.
Table III shows the results of the baseline configura-tion and wczkload as the data amounts to be transferred was
increased. System timel exceeds the system time of method 1
65
'Si
MINUOIS13.6 - B B B B B B- B B B
- A A A A A A A A12.2
- I
10.8
- II- I
- I- I
9.4I
8.0+ C C C C C C C C C
0.0 2.0 4.0 6.0 8.C 9.0 IARRIVAL RATES
A: System time for current methodB: System t$me1 (to the terminal) for alernate: System t.me1 (o the cpa) for floernate
1: 9.6 9.19 5 10 2 63 CUSTOMERS PER HOUr2 : 17.6e, 1 .66, 51 9 i4~ 10 CUSTOMERS PER HOUR17.68V 17.68, 10.11, 5.61 CUSTOMERS PER HOUR4: 7.6817.68. 12.02, 6.20 CUSTOMERS PER HOUR
17.66, 17.68 14.80, 7.47 CUSTOMERS PER HOUR6: 17.68, 17.68, 17.68, 9.04 CUSTOMERS PER HOUR
17.68, 17.68, 17.68 1s.80 CUSTCMERS PER HOUR9 ALL 4 TASIS AT 17.64 CUSTOMERS PR HOUR
Figure 8.8 System times with increasing arrival rates.
between 44 seconds tc 2 minutes and 15 seconds, as the
transfer size increases. The range of differences between
system time2 and the system time of the current method
ranges between 4 tc 10 minutes, as the transfer size
increases. Expected customer loss is again ze:c for bcth
66
i,.;,i,,: .:, , .. ,... ,; , . **.. ; -... -:.:. ..:.: ....:.. -- * . .... . ... - . . . . - . .-
TABLE III -
Increasing data transfer sizesI I
current alternatesyssystem system exptd I
# ctars i ime loss timel time2 loss
e6018. 12.43 0.0 13.17 8.01 0.097920. 12.95 0.0 13.78 8.10 0.0
107712. 13.51 0.0 14.45 8.20 0.0118483. 14.14 0.0 15.18 8.31 0.0130331. 14.83 0.0 16.01 8.45 0.0143364. 15.58 0.0 16.91 8.60 0.0 !157701. 16.41 0.0 17.92 8.77 0.0 I173471. 17.33 0.0 19.06 9.00 0.0190818. 18.34 0.0 20.31 9.24 0.0209900. 19.45 0.0 21.70 9.52 0.0
times are in minutes
cases. Gzarhing these results in figure 8.9 shows the rala-tionships between these system times.
Based on the assumptions of the model, results thus farseem to sugcest there is no apparent danger of customer loss
either at the present workload or as the arrival rates and
data transfqr sizes are increased for the four TASs. The
disadvantage of method 2 is the extra amount cf time
(between 1 to 2 minutes) the intelligence analyst mus-
remain at the terminal to have his output printed immedi-ately at his printer. If the customer does not require the
printcut immediately, there is an advantage because he
completes his work scmewhere between 4 to 10 minutes socner.
This also means there is an extra 4 to 10 minutes during
which ancther interactive search and refinement sessior canbe started. Ccnsidering data fusion efforts sponsored by
the CCINS/rNO, this 4 to 10 minutes is an advantage tc the
TAS because it can receive the data in this shorter amount
67
NINUIES22.0.
BB A- B
-. l + B5A-AI
B A B-I ::- B A AI.15.0. B A
B Aa A
B B A%%,.-.11.5
•-C CC C C C C
8.0,C C C CS------------------------------------------
90000. 120000. 150000. 180000. 210000.Number of Characters Transferredin a 4 TAS's operational environment
A: System time for current methodE: System timel (to the terminal) for alternateC: System time2 (to the cpu) for alternate
Figure 8.9 System times with increasing transfer sizes.
of time and can proceed with the work of data fusion that
tuch sccner.
A valuable asset of any model is its ability to h.lp usanswer the "What if..." questions illuminating potential
problems and benefits. They can aid us in determining scmecourse of action in long-range systems planning. ge have
just looked at the cases where arrival rates and datatransfer amounts were varied within the present COINS-II
environment. Further examination of the changes cf these
68J.l ° .
-. parameters are required as the network environment changes,
in particular as more operational TASs are introduced tc the
network. The next chapter contains the comapartive analysis
of these two methods as one to four TASs are added to
COINS-II.
P. MCDE VALIDATION
The model is validated on the observed reference points.Its results from the baseline configuration is compared to
the measurements from actual performance for average system
time, customer loss and proportion of interactive use. Theempirical data indicated no customer loss due to non-
availability of network ports. The average system time for
all SOLIS users was 12.19 minutes and the proporticn ofinteractiie use was 0.602. Using cuszomer arrival rates and
work profiles from the empirical data, the model predicte4
an expected system time of 12.43 minates, no customer loss
and 0.60 proportion of interactiv usa. Table IV shows this
TABLE IV
Model validation results
modelobserved results7Isystin time(minutes) 12.19 12.43
cus:cuer loss 0.0 0.0
prcpcrticn ofinteractive use 0.60 O.bO
69
comparison. The important underlying assumptions of this
model are the distributions describing customer inter-
arrival times and customer service times. As long as -he
arrival prccess continues to be Poisson, with parameter
lambda, and the service time remains exponential with param-
eter Zu, then the results from the model may be considered
valid.
G. BODRI APPLICATIONS
Cur investigations are based on the demands of fcur TASs
on SOLIS. Network demands on USIS and NUIS were not
included tecause of the lack of empirical data describingthose activities. However, the INS model has baen designed
and implemented to handle these kinds of network services.
As soon as COINS can collect such customer profile infcrma-tion, it can simply be given to the model as input
parameters. No program modification is required.
Although the discussion of model entities were in terms
of HOSTs and TASs and server-TASs, the reader should be
reminded cf their definitions in order to find a mcre
general application of the INS model. HOSTs are pure
servers, and TASs are pure users. Serve=-TASs, on the cther
hand, are a hybrid user and server which both offer services
to and uses services from the network. Hence, when a new
node is added to COINS, it can be categorized as a server,
user, or hybrid. For example, when a gateway between COINS
and some network I is installed, it too can ba classified as
one of the three entities. If the gateway provides twc-way
service of permitting users in network X to access COINS
services and permitting COINS customers to access services
in X, then the gatela7 may be called a hybrid system. Itcan be ±nccrporated into the model as a server-TAS.
70
*, . _ S . . _ . .. . o- . o o o - .. . .. . .- . o .. ,, .... .-.. ,* . .* . .- . * . . -.
. IX. SMO RAUU iNiALrSI
'4. The CCIIS/PHO is actively involved with the installation
of three new TASs. One TAS will be located at Lawrence
Livericre Laboratory in California, and the second one will
be at the State Department in Washington, D.C. A variation
of the basic TS will be installed at DIA and serve as a
gateway between IDHSC and CCINS, permitting IDHSC custcmers
interactive access to COINS. For the purposes cf our
analysis, this gateway is a TAS. It is a source of interac-
tive custcmers to CCINS-II. The COINS/PMO is engaged in
preliminary discussicns with several intelligence crganiza-
tions for installations of a TAS at their sites. In light
of this custcme.- growth over the next several years, we feel
it would be useful to ask the model the "What if ... " ques-
tions. What if we had five operational TASs? ...sixcperaticnal T&Ss? and on up to eight operational TASs.
The methodology was to run the model, establishing a
baseline cf four fully operational TASs. Then fcr each
subsequent run, add a TAS functioning in an operational
* mode. The model was iterated five times, starting with the
configuration of four TiSs and ending with a total cf eightTASs. Table V summarizes the results of these runs. Once
again, system timel exceeds system time of method 1 by abcur1 minute, and system time2 is less than system time of thecurrent method by abCut 4 minutes. Figure 9.1 is the graph
of these results. The interesting result is expectedcustomer loss. There is a jump from a 0% expected lcss withfour cperatcnal lASs to a 4 expected loss with five opera-
tional TASs in the current method. The alternative is stillat a zerc expected loss. For method 1, once there is a
non-zero expected loss at five TASs, expected less increases
about 8.6 percentage points each time a new TKS is added,
71S°
€, - . * - -° ° . . . . . . • . . . . . ° . .. . .*°. .. . . .
.. ]
.3,
TIBL V
Adding a TAS to the network
current alternate
# atem exutd 81te s yst3M exptllotis tise1 tl2 loss
4 12.35 C.000 13.21 8.050 0.05 12.33 0.039 13.24 8.110 0.06 12.24 0.115 13.23 8.130 0.0107 12.30 C.215 13.29 8.190 0.0408 12.30 C.300 13.35 8.220 0.089
times are in minutes
ending with a customer loss rate of 30% for eight TASs. Onthe other hand, in method 2, the first non-zerc expected
customer loss is at six TASs, with about a 4 percertagepoint increase for each additional TAS, ending with a
customer loss rate of 9% for eight TASs. The arrival rate
cf 17.68 customers per hour was used for all TASs. These
expected loss values are plotted in Figure 9.2 . The slcpecf the line describing expected loss for the current systemis 0.0768 while the slope for the alternative system is
0.0217.
Revl.eving these findings in light of the third measure
of prcporticn of interactive work will provide infcrmationcn what propcrticn of the session time is consumed in doing
interactive work. Using the same runs to construct Table V
for the ccmparisons of system times, Table VI wasconstructed showing the proportion of interactive use as a
new TAS is added to each iteration. It can be seen imredi-
ately that fcr method 1. the proportion of interactive useis only at 0.61 while expected customer loss rises to 30%.
72
.'', ,'',", %,... . .,.... ,- . '. '.,-,.' .'. . . . .. ". % '. . " . . .. .. • .. .
89018 EXPECTED AMOUNT CHARACTERS TRANSMITTED
MINUTES14.0
- B B-E B B12.5+ AL
- A A A A
11.0-
9.5"
C C
8.0+ C C C4 +.. ...- 4-.. . . .. + -- - - --.. .. 4 -.. ... .. - --... ..
4.0 5.0 6.0 7.0 8.0 9.01NUMBER OF TASIS
A: SYSTEM TIME (CURRENT)B: SYSTEM TINEI (TO TEMINAL)C: SYSTEM TIME2 (TO THI)
Figure 9.1 System times with increasing TASs.
This is saying while there is only 61% interactive wvck, the
port allccation scheme involved will not be able to service
0.30 cf the customers in an eight TAS environment. In the
alternative method, the proportion of interactive use is 1.0and the expected customer lcss rises :o 9%. The differencein the pzoporticn of use between the two methods occurs
73
..
!I
1 89018 EXPECTED AMOUNT CHARACTERS TRANSMITTED
EXEICTED LOSS0.320
0.24C- A
0.160
- A" " B
0.08C
B
0.0 +2 ---------------------------------.----------.......
4.0 5.0 6.0 7.0 8.0NUMBER OF TASIS
A: CURRENT SYSTEMB: ALTERNATE SYSTEM
BE!EESSION ECtATION FOR A = - 0.326 +0.0768 X1S.D. &BCUT REGRESSION LINE IS 0.02175
BEGEISSICI ECUATION FOR B = - 0.103 +0.0218 X1S.D. ABCUT REGRESSION LINE IS 0.01830
Figure 9.2 Expected loss rate.
because iD method 1, tcth interactive and data transfer wcrk
cccur cui the 15 SOLIS ports while in method 2, only interac-
tive processing occurs on the 14 SOLIS ports with all data
transfers are handled by one high-speed port. The prcpor-
tion cf use remains at 0.61 for method 1 because the
0
74
---
T ABLE V1 - |
Proportion of interactive use
cuarent alternateSroportion proportion
t of .nter- exptd of Inter- exptdUS active use loss active use loss4 0.60 0.00 1.00 0.305 0.61 0.04 1.00 0.006 0.61 0.12 1.00 0.017 0.61 0.22 1.00 0.048 0.61 0.30 1.00 0.09
division cf work in a session is not being varied. The
parameters for the amcunt of time the customer spends in theinteractive portion and the amount of data being requestedfor transfer have nct been changed. The variable being
changed is the number of population sources to SOLIS. This
is being increased fzcm four to eight. The performance of
method 2 is clearly ;referable.The amount of data transferred is the other parameter of
interest. To study the impact of increasing batch demands
on COINS, we started with a base number of TASs, and for
each bass number, iterated through the model while
increasing the transfer demands. The base numbers used werefiv, to eight. Just as we did in the original configuration
of fcur 1ASs, the number of characters to be transferred isincreased from 89,018 to 209900. Appendix C contains thesummarized results cf these runs. The relationships
observed between the system time of the current methcd andsystem timel and system time2 of the alternate approach
generally holds. As the amount of data increases, system
time l increases from about 1 minute to 3 minutes over the
75
16.-
j
system time of method 1. And system time2 ranges between 4.
to 9 minutes less than the system time of the current
system. The interesting statistic is the changes in
expected customer loss. Fcr the current method, each time
the amcunt cf data transferred is incremented, there is a
correspcnding jump in the percentage of expected customer
loss ranging from I to 3 percentage points. This phencena
is not observed in tlhe alternate approach. The percentage
of expected loss remains the same for all increases in
transfer demands. These findings are condensed to one line
entries fcr each netwcrk configuration in Table VII For
example, the second line of Table VII contains the condensed
results fcr five TeSs functioning under larger data trans-
fers ranging from 89,018 to 209,900 characters. The loss
column indicates a range of 4% to 21% expected customer
TABLE VI
Ranges of system times and expected loss
CURRENT ALTERNATE# SYSTEM EXPECTED SYSTEM SYSTE-M EXPECTEDTAS TIME LCSS TIMEt TIME2 LOSS
4 12. 41-19.45 .OC-.00 13.17-21.70 8.01- 9.52 .00-.005 12.33-19.29 .04-.21 13.21-22.39 8.11-10.30 .00-.006 12.24- 19.23 .12-.34 13.23-22.29 8.13-10.29 .01-.017 12.30-19.16 .22-.44 13.29-22.32 8.19-10.30 .04-.048 12.30-19.39 .30-.53 13.35-22.30 8.22-10.21 .09-.09
times are in minutes
loss. Table VIII is the companion to Table VII The table
"- shows the range of proportion of interactive use together
with expected custcuer loss. For the current method,
76
J
..............................
TABLE VIII
Ranges of proportion of interactive use
current alternate0 o~rtion proportion
nter- expected of inrer- expectedTS active use oss active use oss
4, 0.62-0.51 0.00-0.0O0 1.00 0. 00-0.005 0.61-0.50 0.04-0.21 1.00 0.00-0.006 0.61-0.50 0.12-0.34 1.00 0.01-0.017 0.61-0.51 0.22-0.44 1.00 0.04-0.048 0.61-0.50 0.30-0.53 1.00 0.09-0.09
propcrt-cn of use is about 0.61 when the expected number of
characters to transfer is 89,018. As the number of charac-
ters is increased to 209,000, the proportion of use drops to
about 0.50. In metbcd 1, it is quite apparent that as mcre
of the session time is used for data transfers, the price isincreased expected customer loss.
Appendix D has ten sets of graphs for the ten differentdata trarssission aucunts. For each set, there is one graph
showing the relaticnships between the system times and
another graph cosparing expected loss as the number of TASs
is increased. For each expected loss graph, the calculated
regressicn line and error about the calculated line is
shown. Although customer loss also occurs in method 2, the
slope of the expected loss line for method 2 is consistently
smaller tkan the slope of method 1, as seen in the plots in
Appendix D.Close examination of the numbers describing the perfcr-
ance of method 2, reveals that expected customer loss isnot necessarily because of larger data transfers, but rather
the ccnfiguration is approaching its limits in satisfying
77
-,. o - o - .. o - .. . . . . . .... . .. . . . . . . . . .,
the purely interactive demands. This is consistent with curfindings in Table VII, where given a network environment anda certain customer interactive work-profile, no fluctuations
are otserved in expected loss for the alternate methcd, as
increases in the data transfer sizes are made. Referring to
the model of SOLIS as a 2-node network in tandem, providesan explanaticn of this phenomena. Recall that method 2 issimply a reallocaticn of network capacity of method 1,wherety 14 cf the available 15 SOLIS ports are devoted only
to interactive wcrk and all batch work is conducted on cne
port. Increased data transfer sizes only impacts node 2.It was of interest to uncover the kind of situation that
would result in expected customer loss. a closer look wastaken when TUSS is introduced into COINS. The approach
taken was for each increase in data size transfer, the
arrival rate was varied from 2.68 to 17.68 customers perhour. kppendix 2 cortains the results of these runs. They
indicate that a configuration of four operational TASs with
a fifth functioning at a rate of 5.10 customers per hour,
and expected data transfer size of 107,712 characters, thenetwork can expect tc lose 1% of the customers.
We feel the analysis certainly indicates that a reallo-cation of the interactive ports performs better than the
present retbcd. Nevertheless, there are other concerns the
COINS/EPIC must address before deciding on this reallocation
scheme. These lie in the area of implementation. We have
provided the conceptual basis that reallocation is betterand have not looked at the price of iaplementaticn.
although examining implementation costs is beyond the scopeof this thesis, we are compelled to mention the more impor-tant aspects contributing to this cost. In cur view, host
, to hcst prctocol development and network integration are themost serious issues. Converting SOLIS from a 2-stage
facility to a 2-node network in tandem requires prcoessing
78
4.... . . . . .- -.-.
intelligence in both SOLIS and the TASs. This needs to be
carefully specified cn the host as well as on the applica-tion level.
After the issue cf protocols has bean addressed, the
problems of network integration becomes paramount. A well
thought cut transiticn plan must be develcped, wherebynetwork perturbation is kept at an acceptable level. Atransitior generally suggests operating in the old and newsystems in parallel. The COINS/PMO has experience in thisarea, since the network transitioned from a star stcre-and-
forward switch netwczk to one of packet-switching. Dual
services were maintained until all nodes were ready tooperate at the new level.
79
X. c2 D.TS. N U R.Q2...E_.jATjONS
A. SUHAINY
Hithin an ojerational network environment cf fcur
TASs and cne interactive database host (SOLIS), the alter-
nate proposal has a work completion time 4.5 minutes less
than with the current method. This is an advantage if
concern is in freeing an interactive path for a new interac-
tive search and refinement session or getting the data back
to the TAS for fcllcw-on processing of information fusicn or
editing. The alternate proposal takes about a minute lcnger
than the present method to have the data printed at the
terminal site. From a proportion of use standpoint, the
porticn cf the interactive capactiy of method 2 when used,
is completely devoted only to interactive work. However, in
methcd 1, cnly 0.61 cf the interactive capacity is utilized
for interactive work. The reaaining 0.39 is used for file
transfer functions. Despite this fact, the work demands of
four lASs with SOLIS do not indicate any expected customer
loss in the present configuration and certainly not in the
alternate prcposal.
2. .Uzz r cwt_
Customer grcuth was considered as more TASs wereadded. A transfcrm cf SOLIS from a 2-stage service facility
to a 2-node network in tandem exhibits robustness as the
worklcad increases. It is less sensitive to growth than the
current method. As the number of TASs was increased from
four to eight, the completion time to a TAS was abcut 4
minutes less in methcd 2 than in method 1, while completion
80
4%
time t.c a customer terminal was about a minute longer in the
alternate than in the current method. In terms of expected
customer loss and prcportion of use, method 2 displayed an
expected loss range of zero at four TASs to 0.09 at eight
TASs with a propcrticn of use of 1.0. The current approach
exhibited an expected loss range of zero at four TASs to
0.30 at eight TASs with a proportion 3f use of 0.61.
"- 3. 12aia ransf~e Growth
Examining growth in terms of larger file transfers,
the current method was considerably more sensitive to
changes in filesize than the alternative approach. When the
file transfer size is increased from 89,018 to 209,900 char-
acters, completion time to a TAS in method 2 is consistently
4 to 9 minutes shorter than method 1. Completion time tc a
terminal in method 2 ranged from 1 to 3 minutes longer thanthe current method. Similarly, while the proportion of use
remained at 1.0 for the alternate method, this value dropped
from 0.61 to 0.50 fvr the current method for all environ-
ments as the transfer size increased. An expected customer
loss cf C.04 is first noticed in the five TAS environment
for method 1 as transfer size was increased, while an
expected loss of 0.01 is first observed in a six AS
configuration for method 2. For sach network conf4.guration
from five to eight TASs as the data saize is increased, the
expected customer loss in method 1 increases while the loss
remains constant for method 2. For example, in a six TAS
envi:cnment, as data transfer size is increased, the
expected loss in method 1 ranges from 0.12 to 0.34 while
expected loss of method 2 remains constant at 0.01.
81, o,
*4. Feviev
Table IX was prepared to provide a summarized vie9w
of prcpcrticn of use and expected loss when the number of
TASs and data transfer amounts are varied. In the current
*ethcd, the proporticn of use metri1c is sensitive only todata transfer increases. The alternate approach shcws no
- .variation in this variable. Since expected loss is cbserved
* -in both methods as each of the parameters is varied, we can
only ask the questi4on how much better is one from the othsr.
It is quite evident that method 2 i6s substantially more
TABLE XI
condensed comparison Chart
I PROPCIETION OFI NTEBACTIVE USE LOSS
ICURRENI LLTERNATE CURRENT ALTERNATEI As grcwth .61-.61 1.0-1.0 0-30% 0 -9 I(4 tc 8)
IData irowth .61-.50 1.0-1.0 0-53% 0-9%(89 08 toI
stable than the current method.
B. COCUIONS
Jcb specializaticn has frequently been the path to
higher efficiency and better performance as systems grow.
So, it is no great surprise that as customer pcpulation
82
growth and demand for larger data transfers are realized, a
configuraticn whcse underlying philosophy is one of snecial-
ization would perform better than a non-specialized system.
Customer work-profile is a key parametar in the evaluation
just performed. The benefits and advantages of this reallo-
* cation scheme can only be realized if the customer
work-profile remains approximately the same as the expirical
data suggests or if the profile changes in favor of larger
data transfers. In other words, if the profile changes to
where users are spending more time in the interactive mode
than in the batch mode, the new arrangement may not improve
services and in extreme cases will decrease performance.
C. BICCONUNDIIONS
In light of customer population growth alone, scme
consideration should be given to reallocation. However,
when both population and data transfer growth are antici-
pated, we reccmmerd serious consideration of this new
allocaticn scheme. This involves efforts in the development
and evaluation of host and process level protocols and a
carefully designed igplementation plan addressing the rrcb-
lems of operating in a period of transition.
-8
* 83
d, * * . .
EPIRICAL DATA ANALYSIS
TABLE I
SOLIS interactive-only time analysis
sclis interactive time analysis(no bard-copy request)
chi-squared goodness of fit test (exponential)n= 6e mean = 10.13 minutes c qchi- squaredsins. #obs p E(x) statistic
O_ 5 28 .39 26.5 .086-10 13 .24 16.3 .6711-15 13 .14 9.5 1.2916-20 7 .09 6.1 .1321-25 3 .05 3.4 .0526-30 2 .03 2.0 0.0031-35 2 .02 1.4 .26
df = 5 chi-squara = 2.47alpha critical > .25
84
-I"x ,,/, -.' ,- ,- , -t . -' . -. . - - - ..... .,.t.-,-, . . .-. .. *, . , ", , .- . . " ..
TABLE XI
SOLIS interactive time analysis
solis interactive time analysis(with hard-copy request)
chi-squared goodness of fit test (exponential)
n=212 mean = 6.2 minuteschi-squaredsins. #obs p E (x) st atist ic
0- 5 133 .62 131.17 .026-10 43 .21 44.63 .0511-15 19 .09 19.98 .0416-20 7 .04 8.94 .4211-25 4 .02 4.00 0.026-30 4 .01 2.12 1.6631-35 2 .003 .80 1.80df = 5 chi-square =3.99alpha critical > .25
m
85
-1 -.. ,: :-..-........-.... .... . . . .. . . . . .. .• *Z%.
S.Sat TABLE XII
SOLXS data transfer time analysis
chi-squared goodness of fit test (axponential)
n=191 mean = 6.8 minutes chi-squared
ains. #obs p E(x) statistic1- 5 109 .52 99.3 .9476-10 45 .25 47.7 .15211-15 20 .12 22.9 .36716-20 9 .06 11.9 .54021-25 5 .03 5.7 .08526-30 1 .01 1.9 .42631-35 2 .01 1.9 .005
df = 5 chi-square 2.52alpha critical > .25
TABLE XIII
TASI inter-arrival time analysis
cbi-sguared goodness of fit test (exponential)
n=114 mean = 3.393 sinutes chi-squaredmins. #obs p E(x) statfs-ic
0-2 95 .579 100.7 .323-5 15 .24 41.76 .256-8 18 .10 17.40 .029-11 8 .04 6.96 .1512-14 6 .02 3.48 1.8215-17 1 .01 1.74 .3118-20 1 .003 .52 .44
df = 5 chi-square = 3.31alpha critical > .25
86
I p ; ; :; ; . :.: { - -.:: .. ..... ,. . ,.,.. . .. .. . . . .. .
TABLE XI!TAS2 inter-arrival tine analysis
chi-squared goodness of fit test (exponential) p
n=11e mean = 6.!2 minutes ch-qIaemins. #ohs p E(X) statistic
0- 5 58 .53 63.1 .1426-10 36 .214 29.35 1.511-15 9 .11 13.64 1.516:20 7 .05 6.34 .06I21 2r- 3 .02 2.9 0.026-30 2 .01 1.36 .2931-35 2 .005 .63 2.9
*>36 1 .004 .55 .36df - 6 chi-square =7.03alpha critical > .25
TABLE IVTAS3 inter-ar rival time analysis
chisquredgoodness of fit test (exponential)n-e ma 11.76 minutesI
mine. #ohs p E(X) error0- 7 141 .'49 39.9 .0268-15 20 .2149 20.2 0.016-23 8 .126 10.25 .14924-31 6 .064 5.19 .1242-39 3 .032 2.63 .05
U4-147 2 .016 1.33 .33>e1 .014 1.17 .026
df x 5 chi-square z 1.0146alpha critical > .25
87
TABLE ZVI
TIS4 inter-arrival time analysis
chi-squared goodness of fit test (exponential)
n- 29 mean = 22.82 minutes
uins. #obs p S(x) error
0-21 15 .618 17.9 .46922-43 10 .235 6.8 1.5044-65 2 .089 2.6 .1366-e7 2 .034 .98 1.06
df = 2 chi-square = 3.16alpba critical > .05
88
OPERIT!IG INSTRUCTIONS FOR INS BODEL
There are 3 categories of parameter input for the INS
odel. Infcrmation is required describing host(s) charac-
teristics, the netwcrk environment and finally TAS and
server-TAS descripticns.
A. HCST CHBACTERISTICS
For each interactive hcst, the following infcrmation
must be input:
1. proportion of customers doing only iateractive work.That is, those ccstomers that will not _equest a hard-
copy output.
FCBflAT: real. For example: .24
2. distribution describing the interactive-only session
time and the parameters for that distribution. This is a
series of 3 fields.
field 1: must be integer value of 1 or 2 cr 3, depending
on tke distributicn.,* 1 - uniform
,* 2 = ncrual
3 z exponential
F MTCET: integer. For example 3
field 2 and 3: Farameters for the distribution and they
must he in minutes.
if uniform: n1 and n2, where distribution is unifcrm
between nl and z2.
PCBEAT: real. For example 5.2 10.3
89
° : - ... . . . . . . . . . .." . .. . .?" - -" " . . % %.. " .... . -.,% % . ..
if normal: n1 and n2, where n1 is the mean and n2 is -hestandard deviation.
FCRHAT: real. For example 6.8 2.5
if expcnential: nl, where n1 is the mean. Note for thiscase, scme number must be input for n2 even though thatis meaningless. This is because of a minor inflexibilityin the program structure.
PCHMhT: real. For example 6.2
3. distribution, describing the amount of charactersrequested in a hard-copy command, and the parameters forthat distribution. This too is a series of 3 fields.Their formats are as those described above.
4. distribution describing the interactive time when a
hard-ccry request is made, and the parameters for thatdistribution. The parameters must be in minutes. This
is a series of 3 fields and their formats are exactly asthcse described atove.
5. hi-speed flag: used to indicate whether or not thehost is to be considered as having a hi-speed transfer
facility when the alternate configuration is run.
0 = nc hi-speed facility1 =yes
6. number of ports: this is the number of interactivepcrts the host is cffering to the aetwork.
ECEMAT: integer. For example 15
B. iUTUCIK CONFIGURATION
This is the number of TASs, HOSTs and server-TiSs in thenetwork. The numbers must be input in that order.
PORMAT: integer. For example 4 1 0
90
Comment: if server-TASs are not to be considerai as
offering network services, then the number of server-TASs
Must te zero. However, if it still functions as a TAS, they
should be included in the number of TASs.
C. [IS AND SERVUR-TAS CHARACTERISTICS
The fields and formats describing the TAS and the
server-TAS are exactly the same. TAS characteristics must
be input first, followed by the server-TAS. This is so
because the program first builds the data structures for the
TASs, then the server-TASs. The unique thing about theserver-TASs is that they also function as HOSTs, hence they
are alsc duplicated in the HOST data structures with an
extra flag-field indicating that they are really a
server-US. This flag-field is for port accounting
purposes. If a custcmer from 1 server-TAS requests access
to a netvcrk resource, its accounting tables as a TAS are
updated to reflect this. However, its accounting tables asa HOSI must also be updated to reflect a busy condition.
The following information must be input:
1. customer arrival rats per hour.FORdIT: real. For example, 17.68
2. rusber of ports available for interactive work.
FORMAT: integer. For example, 24
3. prcpcrtion of users requesting network access.
FORMAT: real. For example .70
4. Ihe next 3 fields describe the proportion of custcmerswanting access to SOLIS, USIS, and NUIS. Because of a minor
shortcoming of the way the program was written, thesenumbers must be cumulative. For example, if there are .80
ig going to SOLIS, .10 going to USIS and .10 going to NUIS,this information must be input as
0.80 .90 1.0
FORMAT: real. For example .80 .90 1.0
91
5. bi-speed flag: used to Indicate whether or not the TAS
or server-AS is to be considered as having a hi-.speedtransfer facility vben the alternate configuration is ruD.
o = no hi-speed facility
1= yes
92
AIUfIRU. _£
TABLES OF RESULTS
The fcllcving four Tables show performance measurements
of a network configuration vith five, six, seven and eight
lAS's as the data transfer amount is increased.
I
TABLE XIVII
Five TIS configuration
current alternate# char. system exptd pro.of system system exp d
time lOSS use timel tIme2 loss89018 12.33 .40 .610 13.24 8.11 .02097920 12.84 .050 .600 13.88 8.24 .020
107712 13.41 .060 .590 14.60 8.39 .020118483 14.04 .070 .570 15.39 8.56 .020 I130331 14.71 .090 .560 16.27 8.76 .0201433E4 15.46 .110 .550 17.24 8.98 .020157731 16.32 .130 .540 18.33 9.24 .020170 81 17.22 .150 .530 19.52 9.53 .020190818 18.18 .300 .520 20.87 9.88 .0202C9SC0 19.29 .340 .510 22.39 10.30 .020
times are in minutes
93
_"_,r ' ' '~~~~~~~. '..'" " ' " "" ' "' " . . " " " . ...
TABLE IVIII
Six TAS configuration
current alternate# char. system ex~td pro.of syszem system exptd
time loss use timel time2 lcss
89018 12.24 .120 .610 13.23 8.13 .01197920 12.75 .130 .600 13.87 8.27 .011107712 13.29 .150 .590 14.58 8.42 .011118483 13.90 .170 .570 15.37 8.59 .011130331 14.59 .200 .560 16.24 8.79 .0111433E4 15.31 .220 .550 17.22 9.03 .0111577C1 16.16 .250 .540 18.30 9.28 .011173471 17.06 .280 .c30 19.50 9.58 .011190818 18.00 .300 .520 20.82 9.91 .0112C9900 19.23 .340 .500 22.29 10.29 .011
times are in minutes
TABLE XIX
Seven TAS configuration
current alternate# char. system exptd pro.of system system exptd
time lcss use t-u91 time2 loss
89C18 12.30 .22 .61 13.29 8.19 .0497920 12.79 .24 .60 13.97 8.33 .04107712 13.33 .26 .59 14.-66 8.49 .04118483 13.98 .28 .57 15.45 8.66 .04130331 14.66 .31 .56 16.33 8.86 .041433E4 15.43 .33 .55 17.29 9.08 .041577/01 16.21 .36 .54 18.37 9.33 .04
1734"1 17.13 .39 .53 19.55 9.61 .04190818 18.23 .42 .51 20.87 9.94 .042C99C0 19.16 .44 .51 22.32 10.30 .04
times are in minutes
94
, , ", " " - .- . . ... . ..
TABLE XX
Right TAS configuration
6Icurrent alternate
# char, system exptd pro.of System system exptdtime IcSS use timel time2 icss
89018 12.30 .30 .61 13.35 8.22 .09
57580 12.85 .33 .60 14.01 8.37 .09107712 13.38 .35 .58 14.72 8.52 .09118483 13.98 .37 .57 15.51 8.69 .09130331 1.69 .40 .56 16.38 8.88 .09143364 15.47 .42 .55 17.34 9.09 .09157701 16.25 .45 .54 18.41 9.33 .09173471 17.14 .47 .53 19 58 9.60 .09190818 18.22 .50 .52 20.86 9.88 .092(9900 19.39 .53 :s5 22.30 10.21 .09
times are in minutes
,95
L:,4
2".1
I,
S -
* 9
S
.5..V -. . . . .
1D-A132 989 RN INVESTIGATION INTO THE COUPLING OF INTERACTIVE AND 2/3BATCH NETWORK SERVICES IN COINS(U) NAVAL POSTGRADUATESCHOOL MONTEREY CA J B KIM JUN 8-
UNCLASSIFIED F/G 912 NLillllEEimEnniE/limE/I/EEiB/IK//E//E/////EEIEII//EE//IE//IEIIEIIEEEEEEEEEIIEEIIIEEIIEE
* '-.-. n-. .- -.- . -.. ..- - - -
9&11 L1.0 JjM58w
11.11 ..
1a 1 1.
"22
111.25 Ii 16
MICROCOPY RESOLUTION TEST CHART* NATIONAL BUREAU OF STANDARDS -1963-A
....-.
S,.°
SYSTER TIRE AND EXPECTED LOSS CHARTS
There are a pair cf graphs for each increment cf data
transfer sizes. Tke first graph shows the relationships
between the syste times of the two methods. The second is
a plct cf the expected loss of the two approaches. For beth
graphs, the x-axis is the numter of TAS's operating in the
netvczk with arrival rate of 17.68 customers per hour. All
tines are in minutes.
o."
L-.
8 IN UT IS
l.4
BBE E
12.5; AA AA
f - -- -
114.0B
~-
9. 5
C C8.0; C C C
4.0 5.0 6.0 7.0 8.0NUMBER OF TAS'S
A: SYSTEM TI'E (CURRENT)B: SYSTEM TISEt (TO TERMINAL)C: SYSTEM TIME, (TO TAS)
Figure D.1 System times vith 89,018 characters.
97
. *' ,, ,.4 . . *~ -. ,-....,-.. -,. - - ," ,-. .- ,."... . .. .. ,.-,. . .. ." . .. -.- - .. . . .. . . .• 'P'"; ,',,.,,-, , , ,, ." .,. ..- ,. ,,.. ... .. ... .. .. . .. .. .. . . . . .*T .. *4. ,. . . ... . . . . . .
-II
0. 2IISETD LOSI
0.240
- A
C.160 I
0.080B
BBI0.0.2 I
4.0 5.0 6.0 7.0 8.0NUMBER OF TASIS
A: CURRENT SYSTEM IB: ALTERNATE SYSTEM2: BOTH SYSTEMS
REGRESSION EQUATION FOR A - - 0.326 .0.0768 X1:S.D. ABOUT REGRESSION LINE IS S = 0.02175 j
REGRESSION EQUATION FOR B a - 0.103 .0.0218 XlS.D. IBOUT REGRESSION LINE IS S 0.01830
figure D.2 Rxpected loss with 89#018 characters.
98
I. ..,. .;.- .. . .-. ..: ... .. . ... -. . . .. .. .... . . . . . . . . . . . ,. . . . ... .
-, . -.4,7I ,. w .2. . . . f.. - *,, -,o. , . .-. . - . . , . .:. . - . , . . .r " - ,. r - : r
-
'f I--97920 EXPECTI AMOUNT CHARACTERS TRANSMITTED
MfINUTES14.0+ E B B B
12A A A A". I12.5_
11.0
9.5
8 .0 -c
'.0 5.0 6.0 7.0 ;.i
NOMBER OF TAS'S
A: SYSTEM TIME (CURRENT)B: SYSTEM TIMEI ITO T!RMINAL)C: SYSTEM TIME2 (TO TAS)
figure D.3 System times with 97,920 characters.
99
"'' S" " - ,'f * . ." %" - ', t . " .. . * -, " • .. " ., ".:: - . .. . .; . .. .
97920 EXPECTED AMOUNT CHARACTERS TRANSMITTED
EXPECTED LOSS0.360
A
O.27C
A
0.180
A
0.090 B
BB
- ~B0.0 2 E
(.0 -.0 6.0 7.0 .0NUMBER OF TASIS
A: CURRENT SYSTEMB: <ERNATE SYSTEM2: BOTH SYSTEMS
DEGRESSICH E8UATION FOR A= - 0.350 +0.0832 X1S.D. ABOUT REGRESSION LINE IS 0.02071
REGRESSICN EQUATION FOR B = - 0.103 +0.0218 X1S.D. ABOUT REGRESSION LINE IS 0.01830
Figure D.4 Zpected loss vith 97,920 characters.
100
Io |
i
107712 EXPECTJL AMOUNT CHARACTERS TRANSMITTED
"'- ; 15.5
-E E B B B
14.0-A
A A A A
12.5
11.0
9.5
- C C
8.0 C
e ... .. .. . . ....... eeeee -------... eee....eee...
4.0 5.0 6.0 7.0 8.0NUMBER OF TIS'S
A: SYSTEM TIME (CURRENTB: SYSTEM TINE1 (TO TERLINAL)C: SYSTEM TIME2 (TO TAS) I
ggure D.5 System times vith 107,712 characters.
101
• " " " , 1. , '.,% '.,%,.,..'.'.'.',.;. .c .... . -2".. , .... ," .." * .• , -.,.' - .. .. .'.'- .. .', ".• " . ..
107712 EXPECTEC AMOUNT CHARACTERS TRANSMITTED
EXPECTED LOSS0.360
0.270- A
C.180
0.090; B
A BB
0.0 +2 B-------------------------- +.---------------------
4.0 5.0 6.0 7.0 8.0NUMBER OF TkSIS
A: CURRENT SYSTEMB: ALTERNATE SYSTEM2: BOTH SYSTEMS
SEGRESSION EGUATION FOR A = - 0.365 +0.0882 X1S.D. ABCUT REGRESSION LINE = 0.01780
REGRESSION ECUATION FOR B = - 0.102 +0.0217 X1S.D. ABCUT REGRESSION LINE = 0.01819
Figure D.6 Expected loss with 107,712 characters.
.
-,' 102
.- ..............
I.I
118483 EZPECTz! AMOUNT CHARACTERS TRANSMITTED
16.0
- B B- ! E B
14.0; A A A A A
12.0
10.0
- C CSC C8.0
S....... .......-------------------------
40 5.0 6.0 7.0 8.0NUBBER OF TASIS
A: SYSTEM TIME (CURRENT)B: SYSTEM TIMEI (TO TERMINAL)C: SYSTEM TIME2 (TO TAS)
Figure D.7 System times vith 118,483 characters.
103
1XPICTEE LOSS0.40 A
0.~30A
- B1 0.0 2 E4.0 5.0 6.0 7.0 8.0
NUMBER OF TAS'S
A: CURRENT SYSTEMB- ALTERNATE SYSTEM2: BOTH SYSTEMS
SEGRESSION ECUATION FOR A 0.302 0.0217 XS.D. AECUT THE REGRESSION LINE =0.01409
Pigaze D.8 Bzptcted loss with 118v483 characters.
N.,
1014
e, ,. "; •. . .
130331 EXPECTED AMOUNT CHARACTERS TRANSMITTED
.9 NJIIS18.0
E B B B16.C; B
-I A A A
12.0A- I
10.0
*- C C C C C
8.0-- -- -5-0- 6.---
4.0 5.0 6.0 7.0 8.0NUMBER OF TASIS
A: SYSTEM TIME (CURRENT)
B: SYSTEM TIBE1 (TO TERMINAL)C: SYSTEM TIME2 (TO TAS)
Figuze D.9 Slstes times with 130v331 characters.
105
EXPECTED LOSS0.40+
0.30+ A
0.20;
0.10
A B
BI0.0 + B
4.0 5.0 6.0 7.0 8.0NUMBER OF TASIS
A: CURRENT SYSTEMB: ALTERNATE SYSTEM
REGRESSICN EQUATION FOR A = - 0.393 +0.0988 glS.D. ABOUT THE REGRESSION LINE = 0.01087
REGRESSICN EQUATION FOR B = - 0.102 +0.0217 X1S.D. AB OT THE REGRESSION LINE = 0.01787
Figure C.10 Expected loss uith 130,331 characters.
106
143364 EXPECTIE AMOUNT CHARACTERS TRANSMITTED
MINUTES18.0
-g 8 8 B
16.0A ,A
14.0
12.0
10. C
-C C CC C
8.04.0 5.0 6.0 7.0 8.0
NUMBER OF TAS'S
A: SYSEM TIE )B: SYSTEM TI-EI TO TBRINAL)C: SYSTEM TIAE2TO TAS)
Figure D.11 System times with 143,364 characters.
107
* .**. . . . . . . . .
.i,.':EjPEC7IED LOSSO.50
! -
- I
0.20
I- I
O.10
AB
I- IB
-A B0.0 + E E I+-,,. ......--- -........ ----........ -----
4.0 5.0 6.0 7.0 8.0
NUMBER OF TAS'S I
A: CURRENT SYSTEM IB: ALTERNATE SYSTEM
REGRESSIOtI EQUATION FOR A = - 0.393 +0.103 XlS.D. ABCOT THE REGRESSION LINE = 0.008702
REGRESSION E"UATION FOR B = - 0.102 +0.0217X1
S.D. ABCOT THE REGRESSION LINE = 0.01819
Figure D.12 Expected loss with 143,364 characters.
.1
"" 108
' ,,"~~~~~~~~~~~~~~~.--.-."..-..'.-...-... .. i. ...- .. ...-.. .... ,. .- •i,
157701 EXPECTED AMOUNT CHARACTERS TRANSMITTED
HINUIES19.0
-B B B B-E
16.5+ A A
11.5
*C C C9.0; C C
*4.0 5.0 6.0 7.0 8.0NUMBER OF TAS'S
A: SYSTEM TIME (CURRENTB: SYSTEM TIME1 (TO T ERMINAL)C: SYSTEM TINE2 (TO TAS)
Figure 0-.13 System tines with 157,,701 characters.
109
, • .. . . . . . . . . . . .- .o . O o . . . .o . . .. . o . w o . . . . . . ~ , .. v . °.
EXPE.C7E LOSS0.50
.0.4 A'-: 0.40
A I
0.30
- A
C.20
A0.10
BBE B
- A B0.0 + -E
4.1 5.0 6.0 7.0 8.0NUMBER OF TASeS
A: CURRENT SYSTEMB: ALTERNATE SYSTEM
REGRESSION IGUATION FOR A = - 0.403 +0.107 X1S.D. AECUT THE REGRESSION LINE = 0.0099
REGRESSION ECU&TION FOR B-- - 0.102 +0.0217X1S.C. AECUT THE REGRESSION LINE =0.0181
Figuze D.14 Expected loss with 157,701 characters.
110
i . . " .o . . . o . . . . . , . -. , . . . .- . . . .° . o . .
173471 EXPECTED AMOUNT CHARACTERS TRANSMITTED
MINUTIS21.5
19.C
-A
165A A A A j16.5
14.0
11.5
- C C C C9.0
4.0 5.0 6.0 7.0 8.0NUMBER OF TASIS
A: SYSTEM TIME (CURRENT)B: SYSTEM TIME1 (TO TERMINAL)C: SYSTEM TIAE2 (TO TAS)
F igure D.15 System times with 173,471 characters.
' 111
4,
.1
EXPECI E LOSS0.5CA
0.40
0 - A
0.20
*- At
0.10; B
-~ BB
0.0 + B E------------------------- ---
4.0 5.0 6.0 7.0 8.0NUMBER OF TAS'S
A: CURRENT SYSTEMB: ALTERNATE SYSTEM
REGRESSION EQUATION FOR A = - 0.398 +0.110 X1S.D. AROUT THE REGRESSION LINE = 0.0148
I REGRESSION IQUATICN FOR B = - 0.103 +0.021 Xl-.D. ABOUT THE REGRESSION LINE = 0.0185
I-_ _ _ __
Figure D.16 Expected loss with 173,471 characters.
112
V.
19C818 EXPECTID AMOUNT CHARACTERS TRANSMITTED
~10 EEB B B
8..A A k A A
15. C
12.0
C C
405.0 6.0 7.0 8.0NUMBER OF TAS'S
k: SYSTEMI TIME (CURR ENT)B: SYSTEM TI1191 (TO TERINAL)C: SYSTES TILIE2 (TO TAS)
Figure D.17 System times with 190,818 characters.
113
IXPECTIC LOSS
0.50
0.40
0.30; A
C.20
0.10- B
0.0+ E --- ...... ------- - ----4.0 5.0 6.0 7.0 8.0
NUMBER OF TAS'SA: CURRENT SYSTEMB: ALTERNATE SYSTEM
BEGRESSION EQUATION FOR A = - 0.398 +0.115 XlS.C. AECUT THE REGRESSION LINE = 0.01880
REGRESSION ELINTION FOR B 0.102 +0.0217 X1S.C. AEJUT THE REGRESSION LINE 0.01819
Figure D.18 Expected loss with 190,818 characters.
114
.. i....---------------------. -- - - - - - - - -
2C9900 EXPECTIC AMOUNT CBA~ACTERS TRANSMITTEDI
MINUTES24.0
-E E B B B
21.0-
15.01
-C c C C C9.0
4 .0 5.0 6.0 7.0 8.0NUMBER OF TAS
A: SYSTEM TIME (CURRENT)C: SYSTEM TIME2 (TO TAS) ALB: SYSTEM TIMEl (TO TERMN)
Figure D.19 System times vith 209v900 characters.
115
II-V
!XPECIED LOSS
0.60A
0.45; A
".. C.15-B
A
- B0.0 EaC 4a -
t4.0 -.0 6.0 7.0 8.0NUMBER OF TkSIS
A: CURRENT SYSTEMB: ALTERNATE SYSTEM
REGRESSION IGUkTION FOR A 0.384 0.117 X1S.D. AECUT THE REGRESSION LINE = 0.02349
REGRESSION EQUATION FOR B = - 0.102 0.0217 X1S.D. AECUT THE REGRESSION LINE = 0.01819
Figure D.20 Expected loss with 209,900 characters.
116
............. ............ .. ..............
FIVE TO EIGHT TAS COIFIGURATION
These Tables are the results of simulation runs of a
five TAS environment, where four TAS's are running in a
fully operational mode and the arrival rate of the fifth TAS
is incriased over each run from 2.68 customers pgr hour to17.68 custcuers per hcur. For each of these arrival rates
for TAS5, runs were made with varying the data transferamount. The lastda ccluan in the Tables refer to the lamtda
cf [1S5. The remaining four TAS's are running at 17.68customers per hour.
TABLE 1I2
5 TAS, 89,018 characters: expected transfer amount I1current alternate
sstem exptd prop. e s stem sstem exptd Ilamkda time lcss cf us timal t.me2 loss
2.68 12.33 .0 .61 13.18 8.05 .05.10 12.33 .006 .61 13.21 8.06 .05.61 12.32 .011 .61 13.21 8.07 .06.20 12.33 .009 .61 13.23 8.08 07.47 12.32 .012 .61 13.22 8.08 .0 !9.04 12.34 .015 .61 13.22 6.07 .014:8 12.33 .031 .61 13.23 8.10 .00117.68 12.33 .039 .61 13.24 8.11 .001
times are in ainuteslamda rate is per hour
117
_1_.
.-. . "
TABLE XXII
5 TAS, 97,920 characters: expected transfer amount
current alternatesystem exptd prop. system system exptd
lambda mIae loss or use timel time2 loss
2.68 12.84 C07 .60 13.81 8.16 .05.10 12.85 .009 .60 13.84 8.18 .05.61 12.84 .014 .60 13.84 8.18 .06.20 12.85 .011 .60 13.87 8.20 .07.47 12.84 .016 .60 13.85 8.20 .09.04 12.85 .019 .60 13.85 8.20 .0
12.02 12.83 .025 .60 13.86 8.21 .014.80 12.82 .038 .60 13.87 8.22 .00117.68 12.84 .049 .60 13.88 8.24 .001
times are in minutesIlambda rate is per hourII
TABLE XXIII
5 TaMg 107,712 characters: expected transfer amount
current alternatesystem exptd ;rop. sys:sm system exp-d
lambda time loss o use tmel t:me2 loss
2.68 13.41 .011 .59 14.51 8.30 .05.10 13.42 .015 .59 14.54 8.32 .05.61 13.40 .018 .59 14.54 8.31 .06.20 13.42 .016 .59 14.57 8.34 .07.47 13.41 .C22 .59 14.55 8.33 .09.04 13.42 .026 .59 14.56 8.34 .0
12.02 13.41 .033 .59 14.56 8.35 .014.80 13.38 .046 .59 14.58 8.37 .00117.68 13.41 .058 .59 14.60 8.39 .001
times are in minuteslambda rate is per hour
118
..... :. ... . . .... . -. . . ? -..., . - .. . . -
-Z "6 __ 2I N ILTQ .F i11V.J %
TABLE XXIV
5 TASv 173,471 characters: expected transfer amount
current alternatesystem exptd Erop. system system exptd
lambda time oss of use timel time2 loss
2.63 17.24 .052 .53 19.34 9.34 .05.10 17.18 .060 .53 19.38 9.36 .05.61 17.25 .066 .53 19.40 9.38 .06.20 17.19 .064 .53 19.44 9.40 .07.4" 17.20 .078 .53 19.40 9.39 .09.04 17.24 .087 .53 19.43 9.41 .0i!12.02 17.16 .104 .53 19.45 9.44 .0
14.80 17.21 .133 .53 19.49 9.49 .017,68 17.22 .151 .53 19.52 9.53 .0
times are in minuteslambda rate, is per hour
TAS, 190,818 characters: expected transfer amount
*current alternatelakasystem exptd Erop. system system excti
2.63 18.18 .060 .52 20.66 9.65 .05.13 18.25 .080 .52 20.74 9.71 .05.61 18.22 .084 .52 20.72 9.70 .06.20 18.19 .082 .52 20.76 9.73 .07.47 18.21 .096 .52 20.72 9.72 .09.04 18.28 .110 .51 20.73 9.71 .0
12.02 18.22 .130 :5j 20.77 9.77 .0I14.80 18.19 .157 52 20.81 9.80 .00117.68 18.18 .177 .52 20.87 9.88 .001times are in minctesI
* lambda rate is per hour
119
EVENT LOGIC DIAGRAMS
This Appendix contains the logic diagrams of the inter-
nally generated events of the INS model. The following list
of terms and definitions are included to aid in the reading
of tte diagrams.
1. THCUEUE: Queue of network requests. There is cne for
every possible combination of TAS and HOST. TASKs are
placed in the appropriate THQUEUE as defined by the
TAS and HOST identifier.
2. TASK: Temporary entity that may belong to a THQUEUE.
3. LQUEUZ: High-speed facility queue. There is one for
every possible combination of TAS and HOST. LTASKs
a-e placed in the appropriate LQUEUE as defined by the
TAS and HOST identifier.
4. LTASK: Temporary entity that may belong to a LQUEUE.
5. NU2: Expected service time for the interactive session
when hard-copy demand "is also requested.
6. NU1: Expected service time for an interactive session
when nc hard-ccpy demand is submitted.
7. N-31: Expected service time for data transfer in the
current method.
8. HU32: Expected service time for the data transfer in
thle alternate method.
9. BETURN: Return to the SINSCRIPT 11.5 timing routine.
120
*., " -'.:,:.--- - -.-. , .. .. .. :: .::: 1,. .. .. : ..- ; ... -: :-; . . ....... ::-: :. : :i; ' . : - : . ._ '. . . _ % ,, -------- ,---------..---..--...-..-- . . .. .. ... .. .
-~~~7 -7- -V.*..~
MAI
initialize model variablesrea parameters-
---------------------------------------------create permanent enti.ties - paths between tas ;1pin newr configuration
-------------- ----------------------------
iterate through mcdel for current method thenSfor alternate method
SI ~initialize random number seeds II schedule custcaer arrival to tasstart simull tionat terminatc rnreusI reset model va leess
IIand nosl
aa
iigure P.1 MAIN.
121
*~q*~~ *t .sch *dule,..usto'.... ........ .o . as
TASx.AERIVAL: custcmer arrival to a tas+- - - - -- - - - - - - - - - - - - ---
Iscbidule the next custoler in amount of timedef ned by the exponential inter-arrival idistribution with parameter 1/lambda
----------------------------+-----------------Icustomer network request I
yes n .o--- --- -- --
2. determine datatase host I I2. determine work profile
I I
available none available
+----------- - ------- 4- ------------
1 SeiZe rescurce I1.if queues permitted12 if hard-copy demand create temporar-yschedule event for task with customer
customer to send attributes:in cudin at offi~e task in appro-ime defined by expo- priate thqueue
.- nential 1istribution 2. if no queues,with parameter xu2 customer is lost to
". if nc hard-copy the systemI -. schedule customer
dgpartur in amtof !time def ined expo-Inential distriButionwith param sul
-- - - - - -- - - - - ---
REIUHN REURj
Figure P.2 TAS ARRIVAL.
122
a.- . . . .
TBDEPIRT: Customer departures for current methcd
and fcr cases when no hard-copy requests are made
r update accounting variables Irelease network resources I
S IIf any cther departures at this same time instantl1. remove this event from the events list -2. update accounting variables I3. release netwcrk resources
)i
1. search through all acn-empty thqueue's for atask that canbe serviced by the availablezescurces.
2. if such a task is found,remove task frcm thqueueseize the netwcrk resourceif no ard-copy demands, tim
sch.dule custcaer departure in amount of timedefined by exponential distribution withp * arameter mul .
if bard-copy demand, . _ .scheule event to send print-cmadt ain amount of time defined by exponentialdistribution with parameter mu2
RET RN
Figure F.3 THDEPkRT.
123
- _ _
USINE: Customer sends print command to TAS
• -
schodulq event at tbe user-tas to handleuser Flint ccumand in the amount of time totransmt command from terminal to cpu
IRET IRN
Pigure PoA USEND.
124
* ' '' ". '. ",. " , - . ** ." -. -. -. - ."-,". . .
UC.ARSIVIL: At TkS to handle print command
state of tas resource
busy Ifree
1. ceatetemprar 1. set tas resourceentity, utask, with I jbusy eet~
2.f2 sakintsrelease tas resourcequeue in amount of time t
process theI
Irequest,
Figure 1.5 gC.ARRIVALe
125
SCARBIVAL: Print command arrives at host
4-L state of host resource
busy [free
-................. -----------------------create tempo;ary I. set host resource
entity, st'ask with busycustomer attributes 2 2. schedule event to2. file stask in host release host rssourcequeue in amount of t-me toIrepare data forea nsm i ssi1on.
------------- 4-------------------.. -
EIURN REIURN
Figure F.6 SC.ARRIVAL.
126
. 4, . - ' ., . , - - . - , , . ,, ., , . . .. . . - .. : .-- , .- . .. . . - .. , . ..
! Ii
SC.DEIART: host has completed preparation ofdata fcr transfer
set host resource free
method !
current I alternate4.-----------.------------------------I •I
schedule a comletion of data | release networktIansfe: in am% cf tile de- I I resourcesfined by the expcnent4.al dis- I Itributicn with paraseter 2u31 I I I
* ------------------------------------. 4---------------4.
•---- Ila I-.....-4--, I
.--- - ------------------------------Istate of host I Istate of high speed facilitylqueue ! I I
4.-----------4 4-----------------------------------
busy I free
1 reate temporary i1 . seize high speedItast with cus- resourcetomer attritutes 2. schedule comgle-2. file Itask in tion of data x e- inIlqueue amt of time lefinedby th-p exponential
dlstribut-on withparamater mu32
------------ --- --- ask -
.z2 I IzI I IzI I4.--- . __-__-__ __ ___ __ ___ __-___-
Figure F.7 SC.D3PART.
127
1 ,. ' ., . . - - .,, ' .. , .. . . . . . - . .. . - .. ..
S'e
SC.DEPAIRT (continuation 1)
Izl
+ . .. .. .. + . .. . . ..state of thqueue I+ I
not empty I empty---------------------- -----------------------------------
I. remove eaoliest task in thqueue that IIcan te serv ceda by available resources2: qeize network resources3.i.f nc hard-copy demands,
schedule customer departure in amt iof time defined ty exponentialdistribution vith parameter mul
4-.if hard-copy demands, . 1schdule event tc send print commandto tas in amount of t .me defined byexpcnential distribution vithp alameter mu2.
*- -- - --- e e e e e e e e e e e e e e e e
Ia II a I---- +-i-+- I
Pigure 1.8 SC.DIP&ET (continuation 1).
128
* * C C*
SC.DIEABT (continuation 2)
Iz2,-----+
not empty I emptyi II1. remuve earliest stask from queue2. set host resource busyZ. schedult this event to release host !rescurce n amount of time toprepare data for transfer.
----- ------------ ----------------- +
RE2IU RETUIN
Figure F.9 SC.DEPABT (continuation 2).
129
*~ .* .. . . . .. .~ .' . ..
LTBDEFIST: hard-copy transfer completed
gupdate accounting variables I
- -.....-----------
I method g
currents alternate------------------------ -- 4.-
lichedule customer state of high speed queue
-- - not empty I empty
1. remove earliest ltask fromI Iquee 02. schedule this event in amt of ,time jefined by exponential Id str iutcn with parameter mu32
3. destroy Itisk Ia -- n4.-------------+
if an cther transfers completed at this tie1. reove event from events-list2. update account+-ng variables3 fcr alternate case,
If state of hi-speed queue not-empty,rescve earliest ltast from I ueieschedule this event in ant of time definedby expcn ntial distribution with parameter mu32destrci: ita sk ..
4. for current case, schedule a customerdeparture
figure P. 10 LTHDEP&RT.
130
" ,4 . - , -. ',- - ' - . _' -, . .. - _. , - -: _ . .' '. ' ,- , - . . . , . .. . " . . . . ' ' . '
CICSIUG: Halts customer arrivals to the TASs
Cancel the customer arrival to the TAS
RET IRN
Figure F.11 CLOSING.
131
S3
131"
,-,, , '.. . - ' .. ' S, . . S, , - S 5).. . , . . . . . .. . . '.. . . . 5. - . . .- . - -
~~~~~~~~~~~~~~~~~. ....... .. ... . ,..... ...... ,.... ...... . .... . : ..u ,. ... .,, ,;.,
:.,
INS PROGRIN LISTING
This 1p endix cortains the program listing for the simu-
lation ucdel and the Job Control Language statements that
were used to run tha simulation at the W. R. Church ComFuter
Center.
.13
.'
• . 132
*44Q
*14 64*H E2
* 1-4 E-
M E-4 E- *n
mm pa0. %H
04 1114 *W :04 02
*000 di Ad4 91as== tn *m (nb ..
*4a Wa H4. HH =0a% . r@11 Oda C-4 U)s 4H M CD
*h 0.. ocm 4c-9cm E4 44 P Da
*01A4: Do 0.4 i I~ I
*9-04-4* 4 W4raJWa) 2 o 422 A14 2d0W
E--4 222 #J 4 aE4 4H *40.4 * M H2W-4 2 0 WE- 4 V)%=hd *020 L-4* mu) m tn 0 oq Q4 " .o bm W
Od) w 0 * tn =a - M E4wJl 14. a4 * . 9. * 4 4a w n nCo l 0 *E.c
ad dZdw1 * tn W Ird0 42 ow Ww=WQQ5 4 4 E- 1- 0'-. ; *-42AP As * u.4l4H2~ tnQ q2vOJ0 5W v
CW *04""P4 * 04 M 2CDC 4G d.q d* cn 14E04a -; = P"u=~ m (MN *QM~ L4f)4rw Ad 040 p %0S -4b14 =3 6 o(%A * H3 * 2-% mowo4 W 10464"3M H ) E4 4-04C N
*4 bw 41n w-U 22 e E4 U)H.m w in s42- w o4tU) ao
*14H~ Lr) * U) Q4 IQ )4 14oc 0'2 4 b mci 2 cmi o dr
4-0. c* MuTh * Cfl.J044 W I(nFH3-a =1o4 &a14 9-4 0C41 *44h bo M'14 22 W0- U)4d~o W :.
V 41 4Z4.9N **o bU I~UCEC )g4l9UIJ4 .#=~C~~Oc~d *4k * ("1)4 mU) Wiag 04 Q~ = m a3Jafl.W4an adU)"Yoli( *cu *~ 01W W W W... U) ulch4 22 CA 0L
03C * HE-M * ~ MM M :-4 :i1g03 M- 4-1G02"0* 140 ma' . 0iU-4" W 4 Cl)4.0 ) U 44 0 ot
6UQ **O ..9hamma 0WWI-m4ftW0 9-M ~ ~ 2 ) C(44 4FJ
.. 0~* u4o * G)U120)o14~042 PQ*4)0* 44Q000 44
~gg ~~~aa 4 F'~z4a~,~a 1 33 I 94 l )2 l .
;4 c
cncn -3cn
W40 04 M-
0404 Nq H
Nb) CDWHo )c
E-4& Doem14
m 8H m tnHH E-4 4CO4
MU)tflU) t04 4
E-4 tnC c3 c
E-4 % H
04 HU AdH q H Hzo
E-4 :mN01 %H C3 H- O
do 04 L) =0L 4 004 W OJ1 E4 w-H
d4 N4 9 H =.-4U) c1E-4 E14
0 0 E41k 0 o0 z
Pao 0 N w %~z PQ~ Na U aWnU
QN Q V; 04 H 3cO 14a 0
:00 4 W& U4 OHU 14 tnE ~ H )-4. En %-lI tn" Q E4 1-4
U ~ I %c U) UE4 W IW 1L1 4E- WOCW= l H 04 tN w a v414 0 4 0! 0 4 0%1
U).LU M4H Hl OQU04c 0.4 2S 4U f a( -%PoNa NUt % 1 U) 1O ONZ ==An " 6M =0 0.4
2Q 2 I#2U2 4)E-44E 4to 04 a-)o % 8 t 004cns-H sN H E-0 H4414- HH H Nd SH*4 0 N 4
H=44 O 4I b-4 204 2e HCnE -I024basm3ft4w ul1O 4 *E4MN~ QamN H 2 O~M AU~ lb 04=m
WA4 0000 H4 *-0 4 "H r4 *2 W E-4EM W
M4PO w3 04WCUO WC M Oc ImU
pp 4 O 0-O44 4 44 E-4)44s44
E-4 4 pa
k 134
En mga
w $4aca
H w4 H
4 4 ow A"Hn po a -
W 0 E4 "im paE- Hg C
J 09: 44 H-.cn00H HnasE-46*1 c4 E- 0
F4 4A cn 84i.04p -4 f
~3H in cn eno40 asG E-0i 04 Oo
0 I4 C -4 me 02E4 E-4=m %H 0Z 4 (n ZnH .8E4 en %. 0
in wa CzHi-4 0 0ene P n 0 23 E404 =04 404 8-4 64 PQ E4
o~ Kf.en 0 ow tn: W4 E-4 U &40 Dotn 64 Aam44 4E84 % v" raE-4 ra W~ 8O 4 -4 tn o H I
'1w 04 4 4 84 E0PQ H(n CA ad ocn E 40.
'-4 4.4 * 412 one
W4 &'-1 0 %14 cn 84ui
I* enw Ote 8 H H I I H H H om n ' ' 4 8 ~H H -0 *m U04 0A4 E- 4 od =024 = ar~4 =
E4 W4~ m.[08mc4 in miE V H %E-8. 4
e n qw~- 44. U..e....H C.4n8sco o cn % %14E- Amt -en3 484 090 =onmoI-e %0M4 me L- Em4 *cn E4~~ *eno
09 '-4lW M H k440h40 4444 g E4 4 4 0had -4 JJ'H ix %*E4U 0
W44 134 1 4 534X4 p
4.ot.r~j b4:1 J4 > WP 2 0
135
paa
(n 44
-n o4s4 cc
W44 go) 44) WU
04 0.
o4 P13 0 H
rz-4 WH 002U U) 4PQO 4
W4 4 04U
03 E4t 4 n
E4(i W H 04 4E-0H H rl n 4 *d U) H
EI 4 540 0U)win0tna U) 5 0 m ni
4 4 U)) w - d3.44 E4 e b 0
=aq U) vi9 4 l -4 E-404( we (4 4 U) enE s) 4k -iPQ " U E-4 E4 4 oc i
E- 04 00 m4 U)Q 09 44 4
04 04 U) lbD E- .12) -4 b 4 5 44 4 HH H I
0 0Qa Hl 3C4 4 HE 4L-E4 4HH 0 0 0C- 009 =b m so54i- En ad.~ PC Ow3 ad~4 4 44dd 4
a . W=1.141I AC %W e ;l41j E-4 JSUM E ))544U- % %a F-&a .-o .Ia.IUU)U2U))U)U)
m4'"U4 4 U*4 .. ~ 43P W P -4 -4 A 04 O-0 a 04H5.OBC- = 0 0402 4 a 4 a 4 4k-4 acdc 4
%2I 004 I313 V) In C 14 &4 00 42 0
E-4=~ ~~~~S *"4-4a% , ww 4=t iEmv 4P s~innnnn
WaaWE-402MM~n njWQM E4"VM -44 0AM - - Ut d0 4olO
wpau-9 u =4- " cow (" 4Esc aa o oP )v D c -
mutnamag n;mnpl& D mc136a
paum :0 A .w . W W uP
H E4 H H 1F4 6 E4 di
W Iq 04 mfl W 0DO E- - 4 *CflUvl"44Hd LbC) aD A94k4oIq b 04 " W=
.4 c V im 04= Cd4 WUW AC -4U 0 W heA 94=w4.44 adg on No ao) fc9 4 0 00 wp M 390
04 W4H -4~ Ow =m 0 14411 M L -4 H- m% U o" -4<1 di o4 ce4 a* c h0w
P- * 4 41 J Q2 0 e OPe OheheI'.hee oCdh 0 Mo &4 PLs. n U) p tb wPmPQ 00 E-40 E-400000 00 0 00
104~U do 04 9# n i -4.4 m. 14 44 ).4 H inU
.4 )nH HHW~l49 W
"a 4 14 om= tn -4
DIP)OH"041 no2U 14 -4
0 Modal QdHVl E4CD 194 No-4ctn H-4I2b) w~o4uiHmm 0% %m 4
a H0m) paw- 2 22 m wo-o(U m m -)1l4r &4 &1 OI-4 dg 4.1 .Vm. 0-4 0 wo4mmIA m ~H 1.4 =C5 C40 NOW co W %W
-IHwm)PMI H=00.4 U) WNWmPm 21 Ih3C,.4C %a
0 W41 H) Uw14014 .4144104 01 04=4W 1W0 041U)~~~ HO "1)01 4PmE-) 0H 04 & 51 041J115E1o
E- %0 ~~-4C/ .fo W4 .ww =0w94f
* 2 0*" 0he~w. M W) *42'4O= sC 1-4" W 4~~(0.2 54%a0'. "2m Pm)JH'4WH =amp" 1k) 11
0% oz 4 CO=Od A- mmo0s.I ww) 1kUU09'0 W=E14= N=%Doc mW W6)4Mk 11)1 % 4 ==04& E-4 =E-4 1-4w3 WNW) 1-4-414
14)1 w'4'4C ECI)a -,o inI 2 14 U) =02221k WmMb- m c I Z E4 0H20 (Jo U)4U)14E4[41414U)UWv coo ZEL H UcnDp141.4214p4 u)"o a m * u)4 is 94oim
= wgu.4 "4 -'" .442U1 W&Il44 0404"Q "Q0 an 0x U)om040643 (n 04 %ADPO= tEf4m P C- sI4W AVL 0e1-4WE-4 31[404 s4Q1q4h U Eras-I4 =0494O wO = otgcf 0 9O0
20 4 14U).4 eU) %00.04 a E-4=0164 HE-$ 64 =E4 CD o-2 E-4 0- A 400U)U14 0 E4 4 omooCdo %mmo4c= on Pw(1 "tlU oc.414. n-;4 wi).3-.-11 * owk
CD MONk ME-M= Woc= %m w4411201 o4mow) HH o 4 Ho 2s.)EjccUm 2 2144004 No)1 WO.. @00014E'40U= ME-4H= E-4= E-4CD*=NW .1 40 %= tw
%. HH- M 4-9-lI. E-4s- -4 4bs-I.p. s-C s-I .-I s
222012 11J3&jkrI5 44 4p4b4.0404 P-4 P49)" 4
w~w~w ww " " 414 s14 "- s4
00000000=0=0 4 4C4b4& 4 4& 1 - -4E -
137
"A
31 HH WN W 4 0
4. 0 W " HH H HN W0 14a A E-4 E4 E4E4M
WC, H O -m H4 . . a 14 C403PL i.EA H .4 E4 s.~ -4 E4 W a;WWM .1E4 0 .H 04H. A 0 E
OIEM H C 14 4 0 a N4 m4 P4 Wf N- E-MO4mH 60 =J R4= R4M 1 N4 iz x
Hm" mm QEI4 =a mwA 4 0 H E-4
W 4 ow w4 HP a 4 w 0 n a000d 001 R4 0R4 0h4 Re 01 Ri 0 0e Ad
0 0 0E-4 -4 E1
H H H
EA E- 4 6-4ca0-
Ad '4 '4 %M 0 %lb tb Ac WWO ~ W
00'4M 0 OCD W OND In L CD Cd HH lb H'4020Vt 01 WAD4 04 OD4D ri0 WQ N0WC3 go E-4E41E4 E40N'POS4 00S dN0'CA %'0 ' CD' 044N * oH
IRIHON~ ~ ~ ~ ~~"= 04D 4 E01E0 DII.'m 1
'4 M i P w c g 0420 dig 3C 00 PQ M %4040464 0H
ME40 1do (O04 0 m(4Iaw=W OIN Q~mWM 04.-I
1-4 =0UVIC4 WON ifl1414M [4C-4= Mo= .4=Wf14I NWNww4 4 46-eN E-4WOC2'4 lb(/2C4
CV C-4 -4 U) E-12 54.4 04 04 L=an(vHHW1LIp4ZHm
0 n UeCf AU L-4"4. 0kl M-m~~J'H g~ 4" R e L* s1-4HC 141-
UPAN340HO a0 0D D 9 '4' N HH4H .N 4 H
=E14pam2moo eU3 44m~e1~'.~44pim eHod * U H 'Z4i4 '49~~U'p.u
H~~ SM p
D-.o$ d- 10 4 Po I -e04 104 pot sbq Poo -4 -4 m0 m
14P41464 1414 P4 141" 4" 4A 4"m
"444.4 P4.4 '4 .4.4."4."4 4 ww11 p"
138
- 7 -o'
cnp* Q
CD *n 00000
* * .n 0p-* E*a E44* M*0 -
* 000 *- 1 c M w9 pD .lpO
(n *4 0 * c - j*40 *ww=a 3aa wwww
CN M 2* * 0
*l Ur (1 UoU"W L4Wc m*010 =a WwMMMH 40000M
w *Wo60 4n U) - U)
*a WN 04 -NPQ mo *H W - 0w m
* * S _ 0* * 2 H
*139
q~c.
U) H
0411 tn
t=nt U)m
E-.1 0 a 4
=E.4 H- c
H U) .4
ulE4 in H4
m 0 =r -
= Hdo21 -404 lo C C4vH* E U)c r4 a% 'Ic ~ ~ ~ ~ I .4do W94 5 0 )~w =0w
HI 12m m. U)
MM A ~ o C0U. CD U) H -
W4 WNE-4 di Q$.ULI W4 WAm 4040 M ) .4 =*
an IMM W U) H14 Of U) 4 u) H* U-4; A4 4 W40 E4 )0 C EH O 0
H 4440% 2 24'4b2 0 0;04H ca in How- HO wm 0 Ow
'40'4.40~ ~140
*n *3 S I* O41 x II
* * I u-41 m* II *A = I) U)
I *H IO S U)d I J
I E-.I4 aHU OEO I o* 1 I adl I.4E 20-4= I H
* * I 0=11-1H2 a 1 00* 4 I PWC-AHW OCAg40 I b.4
* * 1 00 2 Z&OE40OI 4 1 = si*0* 1 0142UPknPib1 04 a
* * I Qi 1-kE44n H H H=C41 C 4r nh*i4 I m Pa~ci m oi i wai - e l
12 00-0 0 ONDOOI-I- I C-4 m0-*H* * 00 = C-4 -4 w -f 22 N 1- W I H 0 b
E-4* I =,=14E-IH22240 00 1 3c H41* 1 C*Hm1M0W 0U) a-1 En 0 = c*0* I CA 1 .4 " - CNI 10 =2 E a*2* I 04H )HHW0ARDOI En2 H) 1. ACm
t4Mb C V*4- I *a~~0 4 14HU*04* Q -od- M CIMIIIA ~I E-4 H z E44H* I 440 04 154Uk 14 ( 4 Q- w HnMr
**141mc I- &E-44lb 00 1 -ma Hci mu4.* I H -PC61E-.4Wob 1 04 f4t U) 0X = c
H144 I E402920H001 I E-4 04 m m 'An 209H 1 2t"4A 02O4 3C.DkI % M C) H 0 4' *n-A VH1.4 1 140 Old 4 1 1 14 m r - a z .=40
I m pw2Z r U) U)f 1 042 ol .NI Ua)- 0 U g411 04.4ib -4b. w 4w-2 do f Olt 0I~~~ 1 .44 ifi E4 c 0P 414o
E.4 %* 11 ra If~b ~ IfH 3c 14 ' 0441H* of of~21 l4 go to0 (a) 00 99so 5 II w)4E0'4If M-
P* Il 41 r41 2100p 11 H 42141.44O 4 A* 4 -* 4 Ima 0004.0.02 wi H 4 E U- 11S i $
4 4 1 1 Me- -. 4402.I L2 L*0 a i a E-40m 4 w U
411 I IO~.O~ O4w AAW0m 0 -141 4 4 I go bo . J 4 W4 vi2e-
*4 41 H Im P 4 II E14 U 0 . 1 2-41 -14 I Q. in b==H I012 11
=14 I H". 2-0 01d bb E1404 4 0 to~ c . . - ~ ass.a
41411 I I 0141
MCD in 1(4m 4
tn- lb )04
U) +
312 3c -000c
14mt wa- 4i aU04I U) H4 O n -
1-l4 t 14-A0
di tn + E-4 E4H 4A =C4
m %OA 4 31 0 H- 0 -c3i m 0 E4 0 O00%n0
ON Ow4 041=114 14dU) ad A411 9E4 *c 04 U:C4 P H1 M4Add r Ja441 E-4-40 -k a44% a4
C-4 =)~Ua U) cn sa IN E 4 IQ* f
tno)* lb %EM 4 0 W4 ad E- 1 i414- af.-1 12 ] P 4 53 4 b ago ad
a- * 0 a _11dltnw 1 ixH Ilk E-* f a H dE-4 E-4~'IE~ 1-4H 04P )hi D
* C Dw(ic %c 401 "D Pi4 "Ha E-4 OH 10-wo 4u OEO If 00U Os1" 4~b -
Z;I- C wtn (xp wE Occ1c tn 01414 E4 ~OC 1-e w4 4 mhiol =m) MM M ., t iO
mco . O~fl4 '. %m 0 4 W - -0 io m 1 w-, eOC11-I aaIM a0c4 4 n- E4 QoC 9Ul 4044
U)sE4- a AW HM-I 0) A 4 V- QJU -4-4 Od &4 V.V0M~a H 14 14 4 'Cmag
04 'im"O HU Il Q 0 w4 a.-'- 0 1
cri4114mfm *n =44u) go U) U) CL4h321014 Iwo3~s U U4 U 0 D-4 (jo
as d m dM0 =0 00 a i o~ia OU)U b 14 s-40E; Aa
04 H~
h~IAb3I ~ ) I ~ 1I11 II O0O~14.2)
(1)
0
IIt
'a,. so - "3
0 1 i 4 It! OA
ma Mf 94 tOO-f E4 iM in4 E-4 m A n
3c4 3c 4 0 in4~0a 1 [. ~ 00
9404 0w-U) EAl4 d aww-) 4m4 14 m14= 1-4 E- w* E- HI 9414U 12
(n. QI UI (. 0 .go ) 0A 4U w 0 014A04
If 11 OV- 9- E -4 92I.4 02 C4 Q0 Cmqj W1IN 0 HE-4 14 E-4 H so 4b 1 H' W.Iu' M
U)U)- m)U Dol 30I 1 000 If Q E-4 14
0 * * MU 02 ;74 Had owm HoH QM~m ==WON~ k4H7E14bQ 011-41E41140 OA64 CA
ow 06 04 4 C-I P4 W14140ro4 r 4 Q 4CI~q w E4 414E0 dultft44 t-irJU)14Q
g..40 WJ~ M4 3141
143
L U7- 77=.
* v
C- *
* pa
1**114
*I H
*n *
o* E1 "12
*H4 C3 0lO 9
*0 4 2np 0 -
oi w 01P4*E4IQI
* E h- 4 0-1-* * CD 0 ow roIf
* * n~PI~ PQfl5
* * 41 2, H0
41 * ~i 54 1 144
.44t
*M EticU)
* ,q
* )4
0 Hba * E
H E 94 *
H4 %.4W *
0 =) E4 %0*a I- S.4
0 *n V- WE-4 1 41l* 94
6 '%fm * U)' E2E4actnl -% U) 14~4 Hr
E-4 3c ft % v% %A 0 3 C
in bdd)d~ E- 0 I- * Mw
Hn 4b HH "U 04 bbd04 94O 41waatl Mt -m- -CD E4 0 mpo = oc gm H) 94*02 P l q3003 M 40 4
04 00 ~ ~ mms R.--.. 4 1 0 =f'41 0
94 0 Q WI 94
1-4a cdmlu 14 t-40 11 E--994w
H- 0) w Ha-ImaHac) 41 - bSH4
E41 * a~ CV304E- bdix9449 I= E-~)94 4*441 W 01-.44 4 .44P0U M10Z
41~E-a 0 I 204=~ 9
41 .II. U M''~ 1 94 C14-
NO~9sIb~4 .44- 1 9
'la
*-ft*E-4
* * U)**0 c)
* * 4 E-4
* * ) 0
04 E-0
*1 4 0n in
* * lb
0 E-E-b.. *H* PW 0a tn V)
*pa ad 0En F-4V
*0* C
* * * E-4 0 '49.*4* %W 1400640
c0 WAOE4*6* pq wJ .(
*N U)4 Ac in
*o * ) '4 0M * E-4 4 4 00' U3 c4 )
* "-4 Uct4" 0
* * 4 14fl4 = 104 m4 E14 E.4 0 E-4*C* *4 E- U) 4 ad H. '4 H*
* n * ~ m) 042 'H* c) c 35 mE3 39* * 4P E-4-~4HL?* W~ 04
* * U) 4 go' 161* r *W E-4 4 oJ *- m- * C 41
.* 0 LU4 2m fn) v) w.Jl W () U) w-,414 ~I-4" M 14"4 Q4 go0!
41ma * Q2400.- E4 041 w- -zJ4pa ~ U)04ala ('.04d*1 2 m2EA41~~~ us 2 1 14 Uam414 H ul Am*44
* A41*00 m
II
146
-p -' - -s-p.'~ rr~rwcv r~rrr~~wit;.. ~ ~rr~~rr- ~. -* ~flrl ww~whwv~rY~w~ Fvr
* * I1 If
*a *EI I
** Wa i-b- = I
* * - Ad SI
*H* lot9* * E- W Il* * CAI St
* * a E 4 14* * If IS~l
*0* 04 flowIS*2* E4c. 110
CD * * -4 i 3ci*2* 46 U I- Obl
*H=*04 0.11PIU)IE41I
M0 E-4 Ifw - 111
* * 0-M O ItS -0* n41* 0 -0 If in2 4 N 4 I o-
*0*mi ECI4 if49
ad mc.i- 0 S1I. of II
*34 H4do I- ) OCFU) a5-411 u0sU4-4A u 044 =o I SI I
* *4 if % .S 1104 i3 w n - If I Q-
*,13 *-& 46N VN 4 I4 "M(
*o * 0n "(2- C Dlit I M
*c *4 64 5It I1N~f 149
(J40 0410 .2; w- d 2 ~U it I*c4w * 144 C'IC UdW E-4W 1149 51
*L *i 2oo WIS Wi wl0 NWI
* * ml 14-It) UISS 1 00 m H 00. 147 R
- --. . -- - -- .Q
ww7.w .rw~1 .. - -- o
ON EA*M * N* * 0 Cl* * N * C
**9 U* *b 0
*4 *0 0 4-*3 *
*r *- ca4 ~
* 41 014 H5t*o* H H 92 Mb6
*M* ad IDOI a
*- 41n 0it4 - H =E.*- 41a 4 Hn
*4 % o p4 v mf~
41 * 0 P~ 4 a
I=44 CoO H NaomHw410=10004*040 a 4 H
I*E11 "t N 9- C4 E-won1 4i W) E-
* a* 0 0 1-40 64-4 "E041N4 N 144 4a N u 4 4
* *~ *a . 04 vat 4 04 ad)r
*t *f iS- 4 40t44 C4 " 004 H
41W 41 pa0 Of64 m4 O1 0 04 0 E-0 H41 104 N 1. 0 "W 0 . 0
4141 Odig .4 4 Z -4 00 14 0
4 (N8
E-%
**
*- *-4*b * * 0*99 * *
E4 0 %0%H Pag NO*l
* 41%3M~n* v
a * IQ SCO
41 t4 M1 -4
* 14* aCA P
*14*
4%*I
* 4*
*0*
*0* in
* *- %Q %*W L
* * m *z 04o- O4*9E-
*- * 4.PCG* P4 W* d * 4 u e"P04 * P2**u * C=040c ww0I6 44 **
* a *4 0 0 *d ra*
*C *M*E- * * OREf i
0** PL
.. 9, *91*0
*14
E-
01
W4*
I-I4LA.
64 ~ ~~~~~ 0 4%%,%o *0 O
W1** *-t rNf4% "WH %= -4O N. % AM4 OPNA dME 4*-w V
era How w wS0Ef~-4O W4At4=M C~ @ aWl4*=4
* C0
pa 0. E4n Q*0
4*-4pa 2n*
4.7 -- P. 7.. . . . - . . - .. . . * . .
E-4F4
**a*4 :044** *1E-
A A E- 1- ZE-4 EH In
E-4* -a o L
*r W.4 0 H
E-44
H04 PA4* E4 W
at =4 0 -
*0o 0m4 4 544 L- (
0%40to*m * * ~ E-4 =I%00 E *-4* noN%0 0 E4Ho%
*,c * * S.,6000-0 c PIC1NW%-.pto w414 L- d m0 ON4*% a 0kop :A- f
*1-4* w;QC OM 26*= 0 4-.@=
WOU UH42* W1 W4*M.*Od 2x * =*O -0CHC - C- 40
04~~ " 5H4-
D* E*4 54 V -
O* H 44 4
4 2 24* * *L
152
ooU) 0H~ 94
U)n
0 0
0 C-4
94
E- 1-4
E-4* 0l
94 C-4 H
H )
E4 ow *
E H lb H0
0n HE4 M - fr
*W "4 ~wa E.* 4* mom %-
-4 b- a = E- t4*
0 4 wtnc E- ad*
owa* go =0 0() P6 2*= 043 E-4 M cH
* E-4b2 H * - k..1 HP4 64. pq ow P4
02.OS3 PL 0Ib P94944 m- q-cE4 H1 94 U)~ v W
0 E- mwe E-= 0 E- "-4000(*4 E- Ml E-4C M II')V
mW V o 06" W2
- 0 9 w4 L-4a Q29k44 asI 2ua Ho ad MaeM 0&
04 ~ 04 U)0e~ 40d H04
2 **l
r 153
S. *0
*
E* *
* ******C* ** pa* *
* *Mn *"
* Admm pa1E*4 l*=0
w H 94H
in *9-4*tn
*0*
*OW0*ci
*1S4
. .. . . . . .*
- - . - -. . - -. -- . . . . . - . . - . . . . . - - . .
E-
*Q A * *-n1* *4 tmC40-* * a A O- d0
* * -*0* 0 n p inH &
* IS* pq2 .c -
* 4* c 4 W 10ME-
Imc* HOW54 0
*40 *D (4 * 0 E-9E
*z * caw 04 at 041 * q 4vEI- H P4
* H 0" = *02 m=*m *5 L-1 HUI 13 E- 4 SP %sL-*241 2c 540 HZE0PM m
*541 13 C (4, 0md PH- ME4 E-4 054= @
*041 13 54scn Ntc a600 0 o 0 -
OMA *041n 00- W- 13 U t* H "a134-[4
W=1"3L a 0r 45 1 1 0 41411 * )4 H- E-3 544 N
Nam34 E43 2 0 aa d* &0 Pa 0 41*04* 4 m * 1.4 HWS. Ci Q =*1.41 44 m 0 V~m as. U) L-44
*0* 1 0 Ol L4=004 0 b" H 4QMO* 01.2 2* mC4 w1 0 4 L 04 204
* 41 ((200 5 U 14 4 ( Im541 41PQtC (90 .. 1'
44 41 41 2 (4.. 04W bHU4 1021 4 4'9400
E-
* *3E- *-%* *w WM
* *1* *4
In E*
0* En 04
E-* 0E44 -
**= EAW -4U n E4C* *n In r a (12o
a*l M *4 -4 0A E*H* 14 -0141-4)_n E41tnH 114 w r ~ En d En
*w *~ 04 q mw ; s94c 4g
E- U U 0 E- G U 0 L64 F 44M IA Ad MO (12 E- - 0t-E4E18 -0 0
H *~ .PQ PL 4wW OWWW
*14 as4 04 E-4 E- -0- -204 0 0rWC
*22 W 4 2 b4 P4 W1 s-4.4
cis) .ra 1. ~ ) 10 4410 0 .
4 *4 O 42 141) OU) 1~ (1 011
*4 E644 cl1z4 (l 14 . ~ 1
* 441 k 4 04 4 1 1 1 41-~ * * ~ b4 h40~4 ) ~ ~4 SNSJiW
15.6
I-4
1b;
a -c
IQ IrI V)
o a 14 l I
4-04 1 H
-'XI H -0 %
4w
C-4U 14C 0 0 -
13 4 6 D * 04)% =4 -.. 144=s lb %wU F= 4 90A4 H =w*nW3C~ U)w ad- CD 4 )
*04b3,Jl0 ON@= $404 9-4 H
AiE14"H&OID4 U101 %us E4U (LW 14141-4 D :6C H~ a-4 f a* 4 E4 CCE4 (n 04 so 4
1-40Ds14moot) EM= 0 m= t
a ON *UD,1A01sJIC 00 00 00 040 0l4
a 0 MO Nb30000bl40 AA,. AA *404oW o NI c COWAo14 04144 E44 ad
NE~-ON * E'~)U))UU)H) 64C- 4 -144 -41-4 U) 4N H14"N0WO..1J..3s.U400 00 000 4 0
t4&@&4 91J 4 M 4 4 k 1 4 4 L-4 114 = 4 LAE- 14 1r454H *MW=MW1HIME34b4b4"44Ma
1-414.10D40000D 41 11 1414084 M 4
!1 "04041 O4104 1b144Jb alc IsJb 1 2 lzJP,
6"1441111444flf U 157
6'I.
(12 E-4
9.~c E4 4in. E- *0 U26
*Q M 2 0 IM a2v -
*a E- IN64 400 -4* * H A 4 h 3 V)c
*~~ ~ in(1 n.
*1-* alH~ n pq4 EA m E4 0*-4* b . 2 = E-4 0 0 =. 0 E
114 E 04 E-4 m*0* 0 ('4 E4 bc H1* d* 11 4 E- 4 0nt IfA 2 00
Il M .so EM HIM.4b 64 64 04*1-6 641-4.. W 1 1114 %C U 1v1f OwO n. 0*m C4 0 so 0 11 100411400 00 0 14.-4 (12
*CD* H of so =5 d= E-24 (n = 11f0 044 **4* Ow(I M64 4 ''0 1-0w 11 t2 (12 E646
* . .. -1104 E-4 64 = E4011 Q. 4 00-4Q " 0%~ 4 1:412- 04 0 =.-PPU E-404 ". 6~
G; 0% pi 4u oipawom *.-mt4m*44 woc.
z l 0 U Ot.aa w0 MM40a40 6464d I
* * m oft pa E-4 CA 1 A 4 H A n pa CA (120 =* * ol pa0a11 Hj H O2 E- I M z C 4444-c r -
*o S" 14 1 IQ M0
*q &10 02
se 0 E2 A
m A 04 P4 t
158
insa
cn *
* Or* * * 5m
04* paumbb* E-44 -4**
* * 02mmo* 0400
* * mix Q
um * * 04
* * 0 H"MH*2* asamID
HU~ E-40
018 04 4 -
w lM-4 E-
g4 loc U
mma
.E- -4$
*~1.4
* * 43 cb
c QO* 0M 0* 0 O -* * t* * 4E-* um* mm
* *o 04a
*O *446 0
* * 1-4 E4ad 01" p 0DA*mx 141HM ua 1. H N4
* 4 4 - 0 040
04 *.-I uI4 _9MO
*W4 M40*1-4 PWow X -
% ~ ~ ~ 0 aw00 u PioL0* 00 oc m
444 14* a .4a1
u160
* * 0* * 2
* * 0
* * U)
c**Um dig
*14 0a
ow* =i 00
I= a
-4*H " QQ 4 n 4
P~.4
ca im
*4* I1r1
4*4
ad W
- * * L -
to* E- 0
0* * n* * 04c Vp
t* 04 0O*H* alb 04
* * ma C0
E-. * 4 *c~
* * H4C0 .
* ( w-
E-4* bwNV
HUiI4 CflHm
Ea *041*641 41
0 m 4141 0 -4 0 a%
00 W4 04 Im 4 IQ414 04V 06
E-4 4U4 U4~
*4,4 U 4
41 41 4 OW4 I162
* * cn
* * 0
94* 2
* MA*
* 4aON
*M* 102
* * C4 as*~~~~- 4 .HL 4b
* n* 04: 22do* 04 Nj"P
*- * M H DO 1w
*~~ * 04
*~~~ *4 a$~4.4* ~ c *on=14 2
* 4 -4 U -00
* p *0 QMOO
04 UH H 1t.3
*7-*
*q ***%0
* 0* * 0a* * S-r -
P * tnm
* * 41* 41 HI=
PA *- 044
* * 0i 4 j
N 0*n NGI
04 tn-
a% E-1 D-4cr,4 0 1114111
anP 4 U=pUa*mmE441oa
* %0 H4 P411w
V) P44 04 e4W CE-on.4 WJ W4 auW
E-1 4 -4.
41464
M *n
4 0
4r4L* * L
* *4*- *
dig* U' 0I
* * aE 04* * (J 0 N
* * m
* * 0
mm * PL CD
pa tnE44
*4 "40-00
* no = 0
MM E-E=00
* ;f) wow O-4
* -*4 in4 13.
* a-b* Q U
-- C. .. c--. , '9 .-..- 9-.. n ~ 9~*
*4a
Au
'as
-* *
* * 0n
* * 0s
* * WC4 0A* *m 0 ~* *H uvb
** 4n*a *n E-4 w414 4 i a
b44- 1* 04
M04S U040L
as 0 014=
41 * E-4 Q M0
* O* H" -JulW*h.4 w wc 41
E-44 00 au)calE-40
*01 00 O
414* 14166
ad a I Ha I0.4 M * as mo I I
*GQ = QC L-4 I) I E-1 CS E- -no u I~3 Ins I moas
U = M 94E-4= so 9
as H 04 41 ad H. E I*~~ E* ai1~ PQ Is=4 )
C4 M H~ I 1 0 E- q E4 1
* * pa uH IZZ I 19.4 nC. *.as * 0 101 H 0 1 = H.. ..E-4* U)c 0 to 03 4 E-.4 4 * * .- I HUD4 = 1 0 4 I I* ~ c ix IhI 11 I E-4 - w in t*-
CA 1 El C) pa5~* * H 2 H1E4, I =Mtn U4 E-4 I M(*t* A2b so 0Jg IpHa =1 p4 do 04 0a*g pa 4QI i tn0.. i i 1Oct20E-4I *4 C-4 C) za I4h gornip * 1
3 C3 Pap I =IIU) f *Q4m p1614 I Um * wo viw I0IhM A 1 ~ (fC so 0I~a m* Q * 4 adE- m I P4 PU2*I E-)44- 0
H P4EI-22 I V v a I H pqmp 1*2* EA1 Hm mi 0 14 Htn tU 0 -t i5#=N94* Hcn 0c., I 6mM I 0 1414414*H* U)..4 C) HU2Z Ia HU C 2 04am t-*24 C)4 1~ 0d "WOO I U U I w 1U) mO41aU)* 41 H4 44 211 12 I ini "H I o U..114*Ana i mmiV0. as U.H ) %, aS 10424 0 0 1 EI2 4MU14 ) DM4 0 WO- Ia. MI4 I tn , I2 O2..1 0 12
* M 14 Q 01~' H 1 1~ E-142 I P14 *.H -4E -41(&I4 cD wD 2(1)2 8n 00 11 t,(1 rmrO.. x 11 00)
2 u 02l Q - nM 41a D 441 =901C)4 H ms*2*c 2b41 2 QPQ E4H 11-44-1 40M DeM E-09= I I0 Od Q30w12 0 eg mw2 *.-2m. 0111*1-4 u~a' M) = 0I LA I "9-12= I woH214m 2 g IPQe4 H HH H4cm I .140 V VV 4 H Pc *@121414 P4040 14 "L4 4 MCM mH1 "wws 21.4 1O HIpoo 1 ~ f HJ I' 0* - MOO 0.4 0)44 H W I IIg 2 64 Ha (4g.I.~
w 4 so Go, so~,*'efll C C ~ ) 4 14~
41 41 * U) 2 02 1 67 ) 11 H H 1 4 fl e . l f h . I 2
* 4 1 2 m~4 U) H 12 4.4 0 0 I 14~.. ..4I~%
,6- -" 7
7.
--, a
S IS S I
- = I I 0 I SE-4 1 60
,== , b I , I 1 =ISMO. I I
14,":I~ o,1 - 5 I.- ): S = I- I I S'
0 toIII- gI , 04) 041 O04 ID I n a log IN S
1-:' 10 4 1 clm %Ir I I C) I1
I M1, I us%1 11 a
14 ION M! $of I 'I
0 1, I-l I,-1- S Oilr,.
C I N I I WsI I
O 1 , 114 1 N 0 z, IIo I E-4 Ias I OW I UO M!. I.1H lop= I H , u I S I
" I I0
I 31-0 Ia-E I S IIII %E I U* NQ 1-4 I I E0I
N 1 o I E-4 E4* E- I w I I
-U IUiI)i, '4W I I m-IS S I
I 1 uS I m 441 0 I a 3I
I 1 010 camsI a 1 of I
U I I ,I I 1-4 0 0 I O
la-lo11S NISH '*~0 5- 1-4 1 co
I 0 0 *N U 13I1
M!- "168E44 a E-
N l O I D *E N11-429 5 a:= a lwUq w o I Zo pa HH f I I A I
H INPaUt I H 0311 0 E0 w lost I aI
o 1 a% w1 :11 '40 H HeH 14 1 US)fm INNaa I 4 0W "4-4 64aW 40 I I owZUUi I U) I b-011 am 14 4.-I 0*n I I m : W
14 14 I 1-1 $ad Ix 0 14f0 1 =I 14IN amO. UC0 CDPaskI 1 0 1 5 05 4
* 1140:1 G, 1-420,.. e 0 dig n I u oU0.I1-002I 1 U 0, - I l It 0, 1O I I )
-r-"I rU) NO 5'wi H to - Q N " " I to" '' "
I o Wo4 ( 0W1" 4 w ' , '; .0 - 1 41 IQIA , 1 ",N14 10041 1* Ofto"01 Nq 3H64 H- 114 N I pq I w =
0 HIWa I ft f bUCD04 CD H E4 14 11 1I %w I co m 1O.I$4014 1 I14 1 04164 11 g (l so E-4 oI I4 1' 9- 1 H 4AmI I I U) 1I tU)OD U) eta CA NL lof 04 OU(I5EA04HOIIH I ANO IND E-4 CD QA-IINHD0 1 4 1 AIU SNH I ad6"mSU s1 534141 + QI,4Eim 4: I IJ I W "l IO*) I 5 54(fl1 I cpocn tn"L,4 E-4 t-4Uc WI&OID () I I04 41-4= I ~E.40 I H I od04 4dP0 0 NH bea-IODU 21 INP4oI Im E4 I a-
04t~ I 1401al OVU)* 4 0H 14 1 4 1 wO.511
0140H0Ul tn m-5004 E- m I
Om, I.I 04 64041 w O A wu N I ,, U) L3 11,1wI I w UA I c0. H A-I v4I iDnih a- 1 NoI
I0 (U I HOI I NI4114 1 4
I 02la1 in wI( INACpaS
*1161101141 HN u I HID Ec I
Q01 310.1 1 w94
1UW41WPL IH 04
I WooI!IIIHS
168
WHO1
40m
oawoa * a0E4 N 11 = H
664(11.4u C20
U) 0HO0404=01
1". i04Oin.a.= o4
ad 0
U)
ma p
169
II E-
1 4 EA CAW I a I thago* I ccI 4t 4M
I W E-H I I ca I
*a *go I PQ Ad 14dd 2 1C I* *04 1 c M I 1 4 I
*o WOW 141 0I0 IQ E00 I I0 9* paIE4 E- 4 as E- Q
* 0404 1h1 iz t .4no I E- M11. II
* - * d 0400 H U) Od 14 * W=0 I
* I* 0 0 gO 404=0 Iq 0 atn 1 1 .4 I* * E- .I 4~aom a I0 P46 AE4 dd 0 14 WO I* * Ic 0 )0.U 0490 14O M 0 1 "ego* * .400asa 1-4 A 4 0 I 0 3 lkw 4 1 E*-ap0044 *d E-4 Io -4H I E- 100 I
I *n 044C. g U0230 0 M 0EU 1- Ia
Cd M0 P6 HIU m d O I Ma I= pa*0* 0 w..I4owa I U M U PWQ 1 .4 1 0 Q0
*o* =W. OgHD w g I o F4 = z 1 1p o o Ia* * lb 0 E4 004H 1 04 od 8 i 540n E I tn P 1 E- E5 4IQ* M 404.4 5.4 I I O O p I =4 ra I5ia I -0 Wf * o b.4crmso ocoa IU)na E4U I E-s E-4 I 1 .0 4E4 A
E-4* E-4 3CM % W 4M 0- ) 1 3~ Ocn p0o 1 0 04 1 400104 4*0* U) 54054.4H. I 0 i E-I 04 41 M.'~ 02 1Iw 00**0 EM 0U)hIMMI eUPb. 64 E-i as04 I 01 04
* E1* 4 VIj540.4H era I 014W wU)I.0 E- 1I14 =4** * U) %4=04 9 E-0 10E40 MP OI l 0c - I 03CO I 0-P -
-. * * .4045 I= 003111.4 1 U04 OHIOJ I 54 54 %W E4 W 1- i E4 .4
*-4* u) Z bnun=E4 m I0U) cn m I. U) * 4 .QI )ow 4405. =5I~ Pa1 0~ t-4 IOiaU I E-0 11E-
64: E-4 04 64 00 : M U04 40 1 n Pam I t0 U -*1H Ud b.0.4 M 1HL-1 0 U)5 IO. I4 501
*4*~0 Mp4n..1 EU) b 4 H . I w0tu)m Ifl.UI54 M)UMU()5E-0 U)-4 4 0- 1 *o 4p I~ WON .
* 404 ). 505.40.0 et I 0aw mm Ib 3 a aIwan anE~ 41154up* HP46 04 *bd 1I" =ad M 4 I A 0 4 EH HC-4 1 C-H a3 r- PamU)O. IU U)--p 4 U) 1 Mal 0 4 hOw)(fl I E-.-DQE-
* * 0 H4oa .. HI 1 0 nE I C1 9411 154 z54=m4I* U)"J046)..140 I b.-4 M E-4 t I 0l4 Q P4 W 0 04 E-4OU) au I E4 QE -0
ON* PIO * 44F4In 4 1 mHu 04Q" E- 0I. 0 0 14 54 1 40
a$ *M MM M 0I 2 Bu pa01 va= 4l I u" M WAawI = 4HIW * 54 Hm"-IHHH 141 u "4h " S kgla (. " Q I HU.-4Q-4I12 * 0 wpp0AIAQ I Ou 04=0oI HO i . I W404 dg= 9QQMP41al I5454 EAo~n MWa SP3'IMIV00 J0 1I -40HI
* - * Q j~ijQ I 00~ipa e4 I PLUO I= I an E-H* *d9 LM 3OJ 4 U)BI LA.E-4 0 S 0 41
4, I *WU 0 I I I tad
170
I of4
I E-g-
IU) M - 10 161 ad 13 1~~1
0~s 1 coa 0 = I iU3 1 ) W4 64 INN I nE4 1
04 1 as 0 S.1H I E-C-44 '4 6 1 = m U
r3 E-4 .I 0 E-4 6l 14cn I14 Wm6 N I PQ g=) M S -4 1
w4 a0 HomnI= = E4IsoIxo1 o4ow14 0 Vg g) N w4lv
M g E4 .4b H 14 gzU gd = N -
Pa gE4 j I. N g- m H) U) g1 4 g N04 gM g-g HANIg 0 E- '4 g mi gI
in I 4 HI W'W4 4 W4 11 0. 61 Hg
0 gE.4 as co)NN.Ig I = Ad3U gIm I oo4= I HZ of~ (3 IM 0Im
Amb w o Q0P4g N N3 14 14 in # I NF
64% gEM-I I a4 E- As C4 ICg-41 4 g10s~ 1O g.~ '4 wg gm = rdw = WE4Im
-4 * 4 0E4H 1 0 H 0 0 4 LI 4 g I
640 N l3 E-4 0 9 1 '4 4 04 E4 -4 E1)-4 U 0 1 cI== 4 g-4 AQ o'41414 I 1W H '414 )I4O 6= I01 sc a E4 I 014-.414M o6L 0 1 PQ ma61~ 6 E-4
E-0 dII gO0.0 40*t 1 04 W 1 61 hd @a Q I U) 414 I 42'4 11P 4=06 = In CD E -4 W 't- l 460 1
In6 n 3 mr4asce E -4 W 0 =E-4 N I U) I op
E-C3= t E- N Q4W 14O 4 H E-40 O.61IJ I HMacC241 g 0fl 014061 14 0 E-4 3r.4 16) al E- m=A ~ ~ U 1 z m alld IZ C 220)4 110 m J) zn 1 0 1
am1 IN 1 '- 4 mIm 60 0E40 6 C-4 I-I I1 E-'442 H 0N 111 - 04. 1WP61 I '40 14) 661OW 4 OH ")11 0E41 .
us=5 1 .40 En Eqm 14 P4 ~ 44 '4 6140)4 3C 0 (=CGI t"C-4 E-4 1 92 =1 4 C4~ 1 0 66 E-40 HHl .4 as II 6 3 0 1 I '
U13C In 1 3 .094 W I H U)Mq-N 1-4o0 W M C4 1 1-4.s
Op ~ n*bd o4E4Ib g-4 E-41 014 E-il 0 04M O S MO U(Sowu 1 MNON 1 1 W 1M 144 140 ONO% =0I Z
142 on I =2i-0 06 0 =9o OHm E-4 M4'0C24 UgN owED0 CI CMNH OW.E-004 KvE- 4 L4g -4 W L-4E4I W -M I E-0IP4 1O4 0mMW%1 0 = W* AW 3P I= H I w N
14 s I P40114 W 0L-I.. 1 C 94 L1 04 W* U 04 HIA b4E41 X %C4I 1 4W
0~ 60 0EU-u O= I E- 000.4N og Am.A dHg MI
o~Q l UPf0l42) IG )IUZ
In- WHISWOMIUM 0 0 21-INLw
IU 1 E40 E-4 16 04 A 04 1U-4
17.1
&4 .V
go I m I
aUA Ol 14)E4 1 1SO 04 041O 1 DO
E-4 ~ ~ I- X 10
%c4 11 ODU*3
-c .do C4UH IHo 04* 40. Q -
4- A Cl MI 0b4p I014)01 4 ) = =
E4 -4 10 10 m
-0 142 0d CO m40 4 IcD 4 0 HE-O2 0 EA I
WadU4)I- P4 E- D I
14OCI F- 94 WF4 In I
4b P Q 04 asM p
H .. Q i - ft Hf Il -4 I PQQ I
4141.1rd% c14 If 04 Ia E4
04f w E04 14 I
A C 94 I C11 10 44 ad =- H 12 I-ACCIM b-4=E4 4 .- C3 IA P SO 1u I
bdo-41114HOeW m 4=0 401 1 Q- Iino ( nL-E4" " =4=c.0U hi I = I
alJimmal4ap Do pap ae440 100 I I11141 PQ 1141 1N
in m 4= 64 E-4U I FlU) I b00 W D "4 4 O3Q Ila m2 eI
min H 101 P4 aa 4 t o I Ad I-
.4111 W14 =H~ 1.0 IH Q I W4 2 E-4 1 1014
*~~~~~ MU4 L-4Cl44II4 ~ U ~ U(
(1172
-9E.4
E-4
444* * 0 0* * H h
* E-*=* u 4 E* * paw
* A 4 R
* * b 0 0-0.,* *i w N04-,* 44 pq Cp~
*" 4- = -0
* * % e%- -
C4 as tn o tot
E- 4M E- .. ,4M
O*O* 02 .4Oc M -L4L-44 14 M 2 p 4L- 3
4E4 4 0 04E4 0 =do -W0
* O* U E-4 0 - 0 - 1 =-404 w C4 NIE- = E0UE- -*E401 14 04 E-4 A*. * 1-4- U = 0
a.. E-4 011 P4 0. usco L0. nO aE9FE-44 Od4M al U) V)d"$M icn -lu E-0 0 E-4 H Q .O 44 Q MMQj 9qv9 4 x0 4 10M P0Jo Q~. A PW 1) on1 1 4 M "No4 04 0 4 U)a " u m Q Q W4r&E
M. *W (a CD W4 0 " 4
* 4 a o 4w PQ *-..-= 0 W
* 4 ' ''~ 0 ~ ~ 1734
.4-
*v w
* * E-4
* * E-4 an4
* *aPI'. * =*
* *j0 90*044* Ad=* -
*0* AdEACodE
m HQ U E-5 1 .. M
*0* -9 H4 Ei24*M* 14 A44 M = .
E* * H ODD t E-4 b-I
* b4 E44 -414 "WQ
* H4 HcN 4oMOnP WNP*0* ' . ~14
o0 tn 044 m '4 CD
*l *1 24 HN P41QP 0.- .2 Q=04 0.4000=
*N* 4 m4 ~U0 14 " E-4.-u MI-I
* *L 0 CD4
* * I4Cfl eM~u) ~ E-4
* * 1- . ~ =POWI ..
* ~ * woH)M 0E4
4. 0I 174
C-4 E4 C)
.44 co*
* * H
*a *n * 14
M E- 3 0 0 0in *, (AU)0
Cg E-1.
F4 C as 0 0ad E-* %) +- + mra
* *n E-~4 - 0% 0 -
0* *- H~
* * F ca um; )14 -
* %wE4 0M14 124E m
*Q* C-a H~ =QPL414 bC EA0 $4 4 HAQ V
4141 cp 0 &4p rH11~ 104 p-I tn N 04O
*-4 cb o a H 1-4 H 0 C 0 2pa4 04 w u A L
*bC4 %b.1 01M 14OD"-44 t. m = =
E141-40 Ht H W 2C le 104 -
*04 Ma (f m H HO
Do4 a = *CC 3C = 2A 2 2U41*4 c0 H0 3CU2
* 41 u1 2 0 II EtIt-4 W54141 U) 1- C0 O
*4 140EU H W L 61 2 m )14V*41 r HSU1 Q 1. 4 M14 64 14
414 H0 4 0 in 0~ np 03 MQPUW411 Win 0P LM E- LA~-
e1 41 an mm 0 HaOUPQ Mo C44) k
41=4 514 E-4 Ubf4D H O 0 m~4 ~ 14 k4
41 0 %Pe4 0 ()n
175
.4
E-
4F
wi 0
J"E-4= H
*C *-4- *
*0 *f * 0* * *W
* * -W
* 4* 3 H4 * od 40 m
*A* U W -sEp* * 00 04C%- -* * i H04Pl
*H* 041 - n caas* * l ad 9 5 0H D
* * mY N- 4 UL 09 s* * NE- U lp0 .43 +
* * M P4 % -E0 0 A4w 4 -( O N H - 1MP'" M4C
44 " tO 4 =4iiu bm M -4iI
1.*4 EN "J Sg dwcl C41 w
$q V4 .CflN4-%Ww asmiE- *C bM *- bd- m0 - 4m aa 1 sm
0 4p41*d-4-w4E4 (42H NN 54
w9 -L 4* OC 3N in E- =NM" C - df
L* 44 E444 i (C~'E4014 n C 440,44(24 = 4 MI .'54 H.414.)-4D4 54C %DPI E-41 N00" oO54- lC CDN64
* *4O0=0C SN"= Q as D Na110 ON
* 44MW- . .r .i .. !4""~b~S r48""" " O'4 -4 4 5 4Q000 sL 19g~~f4 PaPQ 4 P14PEEIIW 014
1 4 4 4 mn1 0o44inU50''D 4 4 4 ma M
* w * W pt LI 54Pkn-w4) P4 5.-EU
.4 ** 0t t44 tA eOC5QECU) f4~S176l
aa
= E- 1.4 w
* -*a 04 E-U. **0 d 944
CA *q C-4 U*40 40 0Ae N
* *-4 M C 4 0 W -
* m~ I= 94 0a 64 C4 L)
* O* m me 0. Am-O- %* * 0 OW E-4 94ra aq%* * E-C- m0 o E-4 W+ VW.l* * 00 M ) goP b* * E-4 0r Z9 94N nH - c
* * .4 mro 040 Ad 0m 4 '4(N4* 0 OWcl)0 (A toE444 940& a*9* 04HPO .t2 0 04-4 0*$* co U 94 30 3c E4=o tt
W* ) 0sqff0 OH H Hcn
4.'* * E-4 M~ 12 pNU 9()A * Ca; m H4 '4 *U) O.
*94*- U)0U pa zip 04 pa PQ =*0* 4b 3 c OHc C) 094= .-
4' .4 M&40 0) 00 0 01 bQ-f(PE4
0 0 0 U b 02 P 2 NA m94H o~p
* M U 0409M40 C4N H HCI),Q E-b *40O6~ 4 E-.4M 09 s 9E4 0 (D E 04 mL4tE4 9402 )
034C(l OM * ca 9 L)04= " *9== Q 2* 04 094 20 0wpptma= cu M PC.0
tw W A 4 t 2Cf 0 vabi .4 C- c E-422- L b
H 4 Cf 0o sow 092 11 * 4pHvM'40Qw%9Q4020n4*= =o E- 0 0#I 0I .40PIM mC. E-) 4.'4* * Cl)b~oE- 0 94N 2(IC11 H 11 .. 903CHHH,-4 400M* E*-4 03CA 0 32 ogC-6 N- H 0 UHHEA-4~ 001.4n 0
'Odd *ao ae0 t; e4 94(.0 0 1 vr 94 D94 (1.to9E4U2U)1*n * f.c0 02 01-4&49tlg t
* d 0 * 2- 2 2 0 IC2 s30 4HU 11 E-4U2U) O4 WE-Z40 514
* p*0a 93 "-laf = 444 UbOc4 H"4l)C) 44I OW U)*a..* "' 1244 94 Cl) M UHPW '49(1249 U0 2I
* * PQ =;q0H Q-4 0* 42 4 h4 2 44 E-
* -* CUJ m0409 ;UQ.. N10Nm opII .22HHII09 Pau Cl)a -4* * PQ 0H20 A0 o49 A9 Ao4P boP4 0HN49 W~'4 ) '
owH H )o4'.~~~~0 w *ww ~ 0 9 2 *'C)'N1E44N 1.44.4 Q4
* QA j19g 0.4 AQIj4w H9(1. H..PW(J (I) C
"49w 2 i 2' * U)~42S *.I177 .
W
dl * H
* * s
* * cn
*- * -4 0 rq
* * H9-4c* ID* * P
* n *. 'x E-4* *a 00 Sk
* * c ) H- 2U c*H* t4C4c -
* * lb H1 A mm* * E4 tU)Ioc L q
*94* 94 HA NE43 00=H
*=* -4 W4 =c=V n 0+
* N M4 E-U)a Wa 14=
*4* H " O n L4o
*4* U) 9 e 09 0-- .2 VA HUmm2a
m0 4 C- WWWOCH E4H HHtE4E-4 O4
* *. &I E4 494H0 L4uw II o 0 -4H-4. P c a'1. E- *l --in =94 ac w U)O9. 14"
* * 0 I H Cn -n 94 N Hw1114=94w*c *rum" = o" IS= MW *0)-t44IIt i3)
* *4 N- PI "-W H 4 Ut-4 E- A 0 MQ U -m.-JU j~j 4 'C
* n * ~ t0 0.45 a 6O teH* * a-4 ul 9 WL - 41 i5 -0td L-0) "L I~oo4 (' tU
* * SN 1-4 Li Wu o 04 0 p w~h 22A-4 da
* 4 OQ 04 U41))2) HH44 C~
* * P. '4ggUI4444C~ 49.(I178N
N.r
4-4
* * 0
** 64
*O W 04 -
w *4 0 ME1u -
* * 1-4 HO w649Wi*~( *4 jC4 11 04 U 4
ad *MPL *) 0. m - OU I* *q 0 4 04 0 E- o
0 *4 0- E- C n-4c
* 4I- * 0r 0 0= U114* 04A 4E4 XD tn==.0E4 E- -rc
* I* C4 CflO H- E- 1 %eO t n -*- *- W PC4 m- 64E4P mOO4i*4 MP * oal Ml tnW-4 11-I% -E1 0
* * oom~ -v paC -4 P - E-4 -*O 0 v-4%*Q
* * lb 0 0- c4 94H M = E4*WIt1 t P
*6.4 6 4 0 %Ql 312 64 pa* :0 W L 0 6 - l 4 E* bd mC 001* U E-116 -4. '4 op tn 1010 %O40 %) U 011 EWl-A
*0* 0= pa 3 va 0+ C V) 04- W.~ 22i w P4 o-40=
*0* b0a064 o..XU) E- DM6 1 U -6042 1E4C2 MPI4 E-4 *)C
E-4 ~4 33,E 0 *HO-4 NI- 040-. 40 * OMJ ll- :A4%Z1-4*64* ") 04OPL I4= ma-6-4ml qmairp "a eqE iiU)) 4
m10 ' 4 -4 00 W46 H'U '4 o1)0 4 " 1-4U 'oam t
*04 m41 as 406J4 -& -i0 0 HWD0M ow=~rc vowUP1U* ~ 0 U) itoaa n*- *&4)U)06C4.#A 4 W/l b4wZV)E4 -
* H4 odi %0 H 642 in 22'!U oc040n MM meo Ap .41 E-4 11OQ~ E4 = H-1=1 =--'4 44l* 64 &4
30* 0 IC6 11 E-4 E-1 V Wd ~ 6 =0k EA P61 an-
41 E- 4 E-04 H. S HO 0 0H WIlC4'4U)H -4U 464E4.-10H2*41 4104 64 49 N4 4 "as* 14440 F-040406I40 .2 64.1.1U
*4 wo- 2l A.w8 E- E8 4'U -4 f WAil H41 1 ol U) U 12 E-404 2C .6 SWQQ 44 0 H4~ 14OO 4
41 ~ M 41l V) U) o-4 sC Q10 U)4 46444 .~H W4 U)4
41 41 c =D W 86 140 WH E-400~-t1O6s~43*41G wr A2E-14 L-406 M '
41 41 P4 '4 paL060 1.
* 4 ~ 6. U)I U)*,~'4 ~ 011798
ca0
03
c
1.4 E
0 W0
pa *4 cU)A
E-4 94fMtn~~i PHb -
140 04 it A) pawA
q o w U) HAQ (n0E%-4 i
ME- EA 04 W4* 0 da Od
PA 004 i 04 14w -02 0 - E- * bZE4
04 ) 1:1 0 a1 - - 1 4 0 D ' W- 09-4
C- 03CP4cl04C~ ON0 9z mw ma *- 1101-0 HU 21-4 bd u U)AO2Q~ai PI
MIIIIEJh; -4 1 t - 4 P6 A 0 - 10 Zm
-- 06S H ~ p O e -Pm P v4 H -4 1-
U) U UO 2 - P6H H cI U%= u 14hH-4iz
H1U9 =E14 11 cn W A -4 0
P61* ME-lU H adu E-4 E0t9 m 0UIH MI "6au)
30 141C64 it Hmpa pi a H o4 14P IIPI04CD14HP" el 04 0hOOPM 3C"id1 "4 MIUI) -IH 4) 9 C t
14 E 24 H mm OOE* 4=40=0a 414 40M
iIE- 4mc E-4* 2H E 4 - W AO04u Ou+ 1.P 1 U=2ne~i 1 d c t4U)W02 IH 2U1411U)wlmc - U)a Old 04 4
mo o4 it -O0404" HO 044~ 1 0
M~0 P4d U3 id& 04.~l 94UU )4 4) ~ U
EAN LA 14J 1-M L4W M"" E24 c b1
H 0 120 Pit 0-4U()'2
:O0419~90 WU - 4.0
OII 0444OUN E-4 0
U) U30 U)13
4cHn 0
N 4~4180
% . ...- * *. - r .
4-in
4'.H
W %
E. E4 H 0O;Jn4 EA
%dU
.4H(4
'.P. E4a 0+ iw. a; E-CD 1-10 401 3301140
adlwW. SE.. E- U-
01 W4 W4
C-4IM 4II=l~iv 29.' E4nE4 %~A 0QW d
E-6 il i)'EE4c0 w l m v in
HH am E40 0r.
.4 CD WH ink-~b UU040-404 UUJU).4C2 *9'-dtE-410 G
uwI ;;wcIQ.44w~a4=J 9 0 9M
c-) in = Up ~ In wa U) H E
Zn W OWHIMHI"IM"II N L4 wPao MI U) 4i.Iwwu P pq N.4 lA
w 1W 04 Pad
181
U):
*Z 44 I-n *- I
poo 4 p w~ E-4
* Q * -14H 1-4 N qH * 20 E- Eld ) ELA
w ma 113 92 L- c ow (*= * HUNn 4 0 0 Claog0 C4 n we tE) n E- 020n09.E4 -
*z *W~ 0 IQ E- E4E-4 0 pa MAO ml EA4 E-4 04 0A M
4 ~ CYH. E- = 0 0 = =z..
* =: ZO04 0d E -
44U 1-402 cI -4 5p =E4 3C Cl U)4 I* M 14- V) 04 w HOC * Cl) ==0
4 * t4 =P4 - 5-4 0- - 0 C4 t a
4 4 640 E4u 0 S4 so A3&48 11:0 044 0* 0 U i WW 4b so a44 25t 4 H 54*54* Mat) 94 0 Ai E-4 3451M 4.Q= 4 HC 204
m4 tfl.. P4 &tb r 054 619 &4 4 U)N tnMN 01404 54C tn 054Wd j 4 bN 0O
6 dd aU H f Cl) pq 0& = 594 0 H f0lA 0%w-4 HO* 4 00 C- 4s 0 4 cn 54 4 W4 aw0 ~ pti t. 0 c.~ c3a,* .2 ON% E- A54 E- U) ENQ=H H E-40FU 0.4 E-4
* * E-4 0 PzHa OH Hl Pa4 0-.. 0 4==E..i eI=* U 4M U U n 54 Q42 Cl)40 5400 .Nfl1 u0 =4
*44 %WW tnu %-I ic &4 "W a P. W #W4R 040*0 ) 0 -1 *5404 H4 U 0P4 w4E4WH4 mow N5*N 4 cq tnH 54 * C p H3 'I, W-MNIM0 0 S
&I= * U)0 =l CfL0 In ~ 0a MOM0 PQhJ(ra 0*24 0 N Cc 0 04 *= - 0 WIt a = C q H 6 &U4 "4"
as *2 4 C, P4) u. 4 P4 N pq " M 0 0.)* a 4545 205 it0 M-4 oCE- 0M0 ~ o~
:w* 04= VI mw 22 04H H II 11 WI ?n1- m 44..*og 04 aW -4 N W E-4 E- & Oil E,4 1-4 ifCl
H E4 a mp- u.N oa = 04 w- "I %4= HIN5C 4 044 HIMUZ(40=NH 04 H4o tn tnt cpc 4 a U
* 4 HN2 4 %E- 0 5404 545 04 fotc un=4. 1* ~ V 4 em~r4 ago4 0 pa14402 0* 4 410 0 E-4 04 I 4~ -Ut)Cl =2- Q4N.44 " W) 2* p* 4= =30 " C,0N4 Pow 041 P4 0.L q a N04U)
*a c *0 W" W1) 054 40 ..~ . .= NI45554rk C* 4 N .4 04 1144 = NO P4 C)21NNN
E- 4 0410 -4 W W W420NI.* C- 4.- X A0 MIMIl-I
Er * 2 XH M e 0 044 l u H AW
m-O H H N m -
182
L
1 0 1
I* 40 P4
Im 1 1 __w
I*C E-4 I HW
IOpsmI .-o f4 1 W141 &A Ww)I E-4 01 W
I ma1 I crm)
1PI IWCDW0CD I14= OWE20I
pa -I v $A E0 9
I . 94= 00I r-mI E-U*ON=
I maw1 0 HCOW~f EnI02 ma I o I404 0 0
a&0 01 ad* %W4 O
I W 641b
-W) 0 - 0II'4H3.M
0 SUI 0 1 0 W 1100062-E4 5 1 W M 5 = I E- EA E4-4OW4 IL IPL 01 1 14 11 0a w U)flUgo 10.RI41 U 14.U 5IE4Ih 11 M 0 E--40012 lcalE4 w 0 L4 O 6.4O0 s tnnP= 04 E-4UI PL, ml ck4 40'm 0 9491 *o1
I4 WE- I H U) 1-0 .. @ao A=-w4
i van I 11 =4 WIM3E4a 0U E- E-4 a%.4)ME-4 ad4 I4 WU) =024 0 0 0 it1 t 0LI44
U24 I m 1 I 0 0 1 E..4 ME-) L-4M=14 ~maU) I01) =dIA O ft PH11400 0 EU)F 004E0 I=Oi
H1 I 1 1 00.4(=00 cr.O0
P0 41124 -940 W so4.40d 004 m 1 1 1 3 '- " 0P4 0AA w W1 U9414- -O
.4 As- I 010 H 0404 9 040(11to 203 511 WWQ II '4(1) i0oc 0 W LAE4 " 94 0
W 042C4 1 pa 04 U) OM 0 pap W 02OPba JA I OW I -140 o' COs4 0 .4 P4 0-lP 404 4WP
004D 1=4 UGN W 4 (n 0
w I H I .- C-0 0 0rA1H 020C1W4 1pI I 0 0S c
183
-2 1- ..
U;S.. I)
if
*E-** *n
* * EPg
a* inU
H *n
W * 4
* * pa 040 o* * E4 01U3 -
* * 0* * C4 C4 0
0 * * QC 0* * U- 00. 214 p 4W p** U)Pu p 0m m
E* * 0 , i ri H
(A* 14 CD 04E4Ha 1 4 2
@5. o41 %,. %ona tnN
414 Ow 043 PQ
0*14 SI)1C ( 3CH 3*C* OW 4~ 14 4U P E
(6-44 ag 0 - 0 0
0141 4 0'"4141 IN H 41 - -
ASON tn 4141 V) 63I W) 0 *3 "
NOW M 2.. p0h 41-41o4
*U) U)U Il u01 wU oI
414 NOW4 E4l &41-4 .QO U414 pap w papa 04 1 U)3U=
411 -)0 1414p4 0 H Ma
4141 ~U) b~ . U 184
E-4
*SO
* *b
* * -*p A
as* LOW
Do* mo f4* t 14 * 14
*q *44 *C--* E-* O -w I
* * U)0p 42E-4* * 0 2- i* * M o 4-
* tn00*
E-4 0% b~aa U) t
*na * 0 9 C-4I
*04* 0 3C 4t
AQ 0 4 0ac
* 4 I= e c -* a0co *k Hfl 4 ' om
* v HO to..I E41-4 H*a ad0 0 400..402M
*M *H Im4H I.Io~Eu)
* - =tn. *14 04IU tn Of
* *2 0 C- 1 4 2 0 140bM
*M L4R N~ 44 141
* *0 OU) 114023l 0=0 9
* * w VMM ONOWMH Q.44
*1Hm E-4 1.e44* * w 21E4*400I410 tn 2
18.5
M 04
-4 * H 5* * 0b 0
* E-Q 1- -
*U* U) 4 mWOW H*0 .4
w4*44ini*4 *- 090U
EH* 0 = 64M
*H4 040=M0QPI U E'40 c
w(4 W4U4 0 0 14H Wa = WE-=
ofJ V) m ImU ~,ijf* * 4 W4 4 E-4H 000
* * ~ 41U)UU) U)I 140.40
044 *90 = E-4 E-4 E-4iE-4 *6 *n -M~ uuMuA~
W 'I-Q E 0 19=2
* 0* Ow 04P4 '4 b4 L-400= *o P1 ov 14u~ cn1 "0
* *4 PCDpa 1-0 14141*~ ~ U) 0 3)W5U 3C
*4 *w 00 ml. 4 4
Q H pa '4 114
1.86
U*
* ***
*o 2*40 ds 40
* 4.mtnnuE4 C4.41 4 4C06
* b4
* ca
* 4 *
*18.
.4:4
* * IQO I (N*0* Q I4
'I* * 01in I
** * I IE-I*0* L5NW 1 ~ 1
A - * * 4 lo
* * mSo I b
tIC* * 4 vam I I ao'4* * to I
* * H g4. 1 W IC411* * b E I p I 0f
* * =I I m 10 iOf = ft a
*1-4* 1 1 94
*94* atH~l I 1n 40 b*D N- 1~ 04 H iH
* 10 0- tin %-Ww13 *a m I 9
*H* , H I totVIH94 tot : C-4 C v"
*.4 H4 0 IN HI UCDIN 004 * 1 H44@1o %
00* s4N I3 1 I VW JEU o u1
** H N.PJ1 11 1104 mc = l
094 t 4 I miI op O n E- m v)E-uP4 c H
.t~~A AQ** a o
* H II Cc 11- M=10%*poa P0
1 w994 O~ t 0024 CD 0W e~) Wm * I t4 tI v Wl W a 00 (N 0414 4
1 11- * E-41 ej~tm H HH .0
* 6 * NHMlo 1 1P. 110 QO s B
*0 IatS U I. mo to 02 a.4 *- * in Hb4 P0 in *4 Q 1
*~~ *Q 0W 1H194 3N 0 *O tJm
* A 10tS0 1 1111" 94 94iEEU S
*~ ~~~~~~~I 8 ~ BS w NZ
.4b
$ * I u p h
11 90 1 H0U.
I * II.. 0 g1 0
*u *N 6004 + I %W 101W
I I MI W 0- 4b 1 1 0% pa H 6-
*4 * m I 6 3 04 0 16 U 0 0 aca I H~g I m 1. 3. 114101 = w a
*u * I99'~ I0 =
I %" I(Iu w I I~ '-4-
A *) - 6..' H - H 1 m HI. E-4' H I
* * 0 4 bI I p a 0 H 0 1. E- IO~ Hil H41 * H 141 * I'. E-39 g.g'u H oP
E* UIE41 0 S 4 4 94 g1 H 4 1 94
W H 2U 1 0. 14U SHI E- 1 1H
U *o M I H OU _ E-4 1 00 '4 )1
Hm 4a 1 041 01 1V- H H~
*H* 0 H I . lI- U- m U. I 04 11T4 C4 C4
= 0 1 H~ ~I0 E =10M 0 3H s 1 64M H
*0* m I al3 104 0 H 6960 1 A 0 0.
m10 H 0 Otl-' U cI H 626H9N4Wpq *.0 oc I~ C) I64g v 0 E-HIE4 U U- 4
*0* o C 3 u H I m4 0 a64t IW _o v _.
*H* %0 1 1 1 4 H PQ I94 .4 H9 H
** 0A s-H Z 1~ 1 9 4 I9' m 0II 0 104 Q v* oi Hg g m 94 03 61 f0 31 r
*04 U m4 14 I a W 9 10P41 .0 4 u
Ha s a 1IH 2 a I I-I Pa 0 2I
004 0 0 0 4 0 4 1 94IfI= 1 13ck 4 -
* s * 2 00 U46 CD I H .H e 0m
* ~ ~ ~ O *% ~. D2(6001 ow 6 0 = -
* * H '0 H -11410- 0 v 4C ol -0 6 C0IE-0 4 + C4 aE- *A 0 ok- I9 mI I m ==4 M= 0 1 II2' 0c rw~ Cp
0* 41 =a (3 V 10 H s 0 0 0 4P 1 060 M= .0 0 .
* km *V 34. HH I l0 0 mm ( 1.. *m w I I ca H U
*o *6 boO 04H P 160 mI. II. In Q14 10 w41 * E-.4 - E-4II ft 0. IIw (2n I O6- -4 H-
1- =01 Ma 02 0W 2(3 (3 0 1" ,9m cj I~ 0 2W0. 4 20
4 4 H1 H. I=6 62 1--I OHC H 46U A I06 HOWA 0W.m '.o solo. I- I V VI
a p0 k4(J IPl4 b - v o 9440*~~~~~9 WOO H H wt14 (J- (.I 41 L64 *' *
* 4 09 -4 06142 00 ZO4.- CD6260 1IZ809
Q ...
4.u
4.n
lbIc v 0
Do u u HH W H H 0S
o- 4 0 0 6
I1 H H
U4 oa 0 3
0% 0 . -% H
m 0 30a %
0 N 0
M4 H H lid0 4 M 0 M0 1
w 0= Ol
4m Q. 1-4 QN t
04 m 0- aE4 E-0NJ ("4 4rm UM c
H4 0 04 1404014 ow a D
01 4D a4 W %2 Ik'
= M1090
"* LIST O REFERENCES
1. Licklide;, J.C.R., Vezza, A., "Applications ofInfcrzaton Networks", Procgedins o e IEEE, Vol.66, Nc. 11, November 1970p. ig-OW-. -
2. Roberts, L.G., essler, B.D "Computer NetwcrkDevelopment to #chieve Resource Sfiaring", AFIPS Conf.E c OS_ z~nq Joint Computer C3nec= o"u
;c..17 vigJi n feren~lWol.I77
3. COINS Project Management Office ZOINS Networkorla aNd .form ance to 1990, b-7----
4. CCINS Pr oject Management Office, Technical Prorosaligr Deeo rme nt of COIN DaaCrea71n.7 -e
5. Defense Advanced Research Projects' Agency (DARPA)RePort P-82-1002-tp, Tech ncal groposal for: Secureni -q(Zd -rn-.Ma 3 v e 2 1 1F -V T -9 1. F e D u a r y 1 9 3 2 . -
6. Informal correspondence from Mr. George f. Hicken,* CCIVS Projct Manager, July 1982.
7. Informal correspondence from Dr. R.L. Wigington,Director, Research and Development, Chemical A stractsService, March, 1983.
8. Dominick, W.E., Penneman, W.D., "Monitoring andEvaluation of On-Line Information System Usage",.igge io _Zrccess-nQ& Management, Vol. 16, No. 1.,
9. Melny k, V " Man-machine Interface: Frustration",~~ Noveuber/Decenber, 1972, pp. 392-401.
10. National Library of 0Medicine, Report No. NM 78-7,d Eva.. .oO the On-LineProcss by McDonald, D._", Santa Monica, Ca.:January 1980.
11. Baker, C.A., Eason, K.D., "An Observational Study ofMa12-Ccputer Interacto Using an Online BibliographicIn mao Rereval s lstem, 2R1111j14 1, Vol. 5,No. ,&rl 1 IS81, pp. 1 1-132.---
191
-AD-A132 989 AN INVESTIGATION INTO THE COUPLING OF INTERACTIVE AND 23BAITCH NETWORK( SERVICES IN COINS(U) NAVAL POSTGRADUATESCHOOL MONTEREY CR J B KIM JUN 83
UNCLASSIFIED F/G 9/2 NL
LEE
1.01
E1.25 11.4 11.6
MICROCOPY RESOLUTION TEST CHARTNATIONAL. BUREAU OF STANDARDS-1963-A
12. Benefrld A.R., Kugel, R. Marcus, RS. "CatalogInformation. and Text as ;ncasors ofelevance",J.. , January 1978, pp. 16-30.
13. Carlisle J.,_ Martin, T.H. Treu, S., "The UserInterfaca for Interactive Bibliographic Searching: AnAnalysis of the Attitudes of Hineteen InfcrmationScientists", J , Sarch/April, 1973, pp. 142-141.
14. Marcus, .S.," User Assistance in BiblicgraphicRetxieval Networks Through a Computer Intermediary",01.r~l~~,; s_ s -en Wn. 141 3ac r
15. Dayton, D.L., Lundeen, J.9. Pollock, J.J., "Automated- Tecbniques for Crnine Search Guidance: A Reviev" 4th.- "aneioa fieeui q, London, 9-T
16. Tedd, 1. "Intelligence in the User's Terminal: A Lookat Current C tions and Possibilities", 5th,. ern Al CalIenforz eeti, Lononn, 8
17. Latcratory fox Information and Decision Systems,.I.T. Cambridge, Ma., - Rep. LIDS-R- 1233_ .I.T. .mh 4 Casoid e, er.a.,._ . Dop . ,_...
18. Meadow, CT., Epstein, B.E., "IndividualizedInstruction for. Data Access",Ist Internaiona1_l_
1H _Aeg=q,,London# 13-TID U6C9 ~r 17;, p.
19. Durkin, K., Egeland, J., Garson, L., Terrant, "AnEx;erient to Stud the Online Use of a Full-TextPr mary journal Da abase", 4th International Online
fcetinaM London,
20. Kiviat P.3., Markovitz, H.M, Villanueva, B.,asc:P;;S;.qqm~(T Ln 1 9,ed. Russel, E.C.,L~a~..Lw*fgeles ca.,utBr 1975.
192
I .' ,', , ," .. • . .
, .-q -. . ' ,''' :.'*..*. " _:' . ...........-.............. i...... ...... .- - J.... .
rj.
INITIAL DISTRIBUTION LIST
No. Copies
1. Defense Technical Information Center 2C12 mornitatiQn.Aexandria, Virginia 22314
2. Superintendent 2ARES: Libra.ry Cede 0142Naval Postqgriuate Schoolfcnterey, Califcrnia 93940
3. SuperintendentATTN: Prof. Norman F. Schneidevind, Code 54SsNaval Fost rau ate SchoolMonterey, Califcrnia 93940
4. Suerintendent 2AT Prof. ff. G. Sovereign, Code 74laval Postraduate School
C Curr cu arnOfficer Code 39laval Pstgrade ate Scoolgonterey, Cali fcruia 93940
6. Directcrp National .Seq urity Agny2ATTN: Hr. George 8. Hc akn gcoils Project management of ficeS600 savage RoadPt. George G. Meade, Md. 20755
7. Chemical Abstract ServicqATTN: Dr. Ronald L. igingtonDirectcr, Research and DevelopmentI.O. Box 3012Columbus, Ohio 43210
8. Massachusetts Institute of Technology 1ATTN: Hr. Richard MarcusLaboratory for Information and Decision SystemsCambridge, Mass. 02139
9. Environment Protcti 1&STYI: Dr. Siney Steiegef lTS401 5 St., S.goashington, D.C. 20460
10. USS Enterprise (CVN-1) 1ATTN: L-. B. FrenchC;eraticns DepartmentIPC San Francisco 96635
11. Ccmandant of the arine CorpsCode CCAATTN: Cant. B. G ahamHead uar edrSates Earine CorpsVasbington, 6 .C. 2038
193
12. dajcz Ta lan AhnComputer nterI.e. scIGongl eul.ong iobong_Gu
13. Hellenic avyITTI: Cdr. asmas CharpantidisGeneral Naval StaffSclargc_Athens* Greece
14. Directc a1ional Security Agency 5|TI: J5B MPa 566
S980 Save Roa20755Ft. George G. Neade, Nd. 20755
ia.
a-9
10
I -I 44
Iw lot 06 %.
A
R6S
POj
