Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
1
Oct 05, 2005Oct 05, 2005 APCC2005APCC2005 11
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
A Flooding MethodA Flooding Methodfor Exchanging Routing Informationfor Exchanging Routing Information
in Powerin Power--Law NetworksLaw Networks
Nobutaka [email protected]
Osaka University, Japan
22Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Overview
BackgroundObjectiveFlooding in power-law networks
Examine the problems of flooding on the Internet-like topology
Proposed methodMechanism and Evaluation
Conclusion and future works
33Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Background(1/2)
Flooding and its problem
Flooding is used to deliver the routing information to entire network
Each node copies a message to all the neighboring nodesSome nodes receive the same duplicated messages via different routes
If the size of a network increases Amount of traffic becomes critical problem [2] by increasing messages Example of flooding and
how the duplication occurs
[2] L. Gao and J. Rexford, “Stable Internet routing without global coordination,” in Proceedings ofACM SIGMETERICS 2000, pp. 307-317, June 2000.
44Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Background(2/2)
Topological characteristic of the Internet
Internet topology follows “Power-law”
Connectivity of nodes exhibits power-law attributes
How the flooding in power-law networks behaves?
Messages tend to concentrate on nodes with many connectionsIt is important to know the feature of Internet-like networks
Many nodes have a few connections to other nodes
A few nodes have many connections to other nodes
Distribution of the number of neighboring nodes in 1000 nodes power-law network.
P(K
) : C
ompl
emen
tary
Cum
ulat
ive
Dis
tribu
tion
Func
tion
0.001
0.01
0.1
1
1 10 100K : number of links
Nodes with 2-4 links occupies 80%
Four nodes with 50-80 links
55Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Clarify a behavior of simple flooding in Power-Law networks
Estimate the amount of control messagesExamine problems with a simple flooding method
Propose a new flooding methodBased on the observation found aboveDecreases the number of control messages
Objective
66Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Evaluation of simple flooding (1/2)
Duplicate messages in Power-Law network
Simulation setupTopology generated by BA model [4] with 1000 nodes (m=m0=2)Failure on a node with many connectionA failure occurred at the highest degree node
Messages concentrate on nodes with many links
[4] A. Barabasi and R. Albert, “Emergence of scaling in random networks,” Science, vol. 286, pp. 509–512, Oct. 1999.
Number of control messagesgenerated by single-node failure
on 1000-node power-law network.
012345678910time (ms)
0200
400600
8001000
0500
100015002000250030003500
cum
ulat
ive
num
ber o
f dup
licat
ed p
acke
ts
012345678910time (ms)
0200
400600
8001000
0500
100015002000250030003500
cum
ulat
ive
num
ber o
f dup
licat
ed p
acke
ts
Nodes with many links
Cum
ulat
ive
num
ber o
f du
plic
ate
mes
sage
s
Messages concentrate on nodes with many connections
time (ms) Nodes with a few links
Flooding in power-law network does not scale well
2
77Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Evaluation of simple flooding (2/2)
Duplicate messages in different networksCompared with random networks
Connections between nodes are setup randomly
Many duplication found in power-law network
Duplication increased rapidly in power-law networksDuplication increased slowly in random networks
A method to reduce messages is required to perform flooding in large power-law networks
Maximum number of duplicated messages dependent on network
size (averaged over 10 experiments)
num
ber o
f dup
licat
e m
essa
ges
(x10
0)
number of nodes
0.1
1
10
100
1000
10000
1000 10000
power-law network
random network
Increased quickly in power-law networks
Increased slowly in random networks
88Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityProposed method (1/3)
Overview of proposed method
(i) Probabilistic floodingEach node relays messages to a part of nodes chosen by a relaying probability pDuplicated messages are reduced at intermediate nodes
Some nodes do not receive messages by its nature But in routing protocols, all nodes have to receive routing information
A node originates flooding
Flooding continues with probability p
!!
Our method consists from two ideas
99Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityProposed method (2/3)
Overview of proposed method(ii) Periodic information
exchange Each node asks its neighborsweather they have a new informationIf new information is found, a node receives it from the neighbor node
Nodes that couldn’t received control messages can get routing information Using this mechanism, we can choose lower relaying probability
A node originates flooding
Flooding continues with probability p
Deliver information to all nodes by periodic exchange
1010Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityProposed method (3/3)
Choosing relaying probability pHow many nodes receive information by periodic exchange?
About 15% of nodes according to the simulation result(not presented here)
Relationship between probability p and the number of nodes received messages
Probability p=0.1 ~ 0.9 (fig.)p=0.6 enables this method to inform 85% of nodes
time (ms)0
5001000150020002500300035004000
0 2 4 6 8 10 12 14 16
Simple flooding
p=0.9p=0.8p=0.7p=0.6p=0.5
p=0.4p=0.3
p=0.2p=0.10
5001000150020002500
0 2 4 6 8 10 12 14 16
p=0.4p=0.3
p=0.2p=0.100 2 4 6 8 10 12 14 16
p=0.4p=0.3
p=0.2p=0.100 2 4 6 8 10 12 14 16
p=0.4p=0.3
p=0.2p=0.1
Num
ber o
f con
trol
mes
sage
s (c
umul
ativ
e)N
umbe
r of n
odes
rece
ived
in
form
atio
n (c
umul
ativ
e) p=0.6p=0.5p=0.4p=0.3
p=0.2 p=0.10
102030405060708090
100
0 2 4 6 8 10 12 14 16
p=0.9p=0.8p=0.7
p=0.3
0102030405060708090
100
0 2 4 6 8 10 12 14 16
Simpleflooding
time (ms)
Delivers to about 85% of nodes
with p=0.6
Decreases about 40% of messages
with p=0.6
The evaluation result of probabilistic flooding with p from 0.1 to 0.9
1111Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityEvaluation of proposed method (1/3)
Simulation experimentsTwo scenarios of flooding
Flooding from a single nodeFlooding from different nodes
Evaluate two indexesThe number of nodes that received messagesAmount of control messages
Compared with two methodsSimple floodingFlooding based on critical probability (p =0.9)
With this probability, 99.9% of nodes receive messages [6]
Other parametersPeriodic exchange is performed at every 5 secondsEvaluated on 1000 nodes power-law network (BA model : m0=m=2)
[6] F. Banaei-Kashani and C. Shahabi, “Criticality–based analysis and design of unstructured peer-to-peer networks as ’complex systems’,” in Proceedings of GP2PC, pp. 22–32, May 2003.
1212Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityEvaluation of proposed method (2/3)
Flooding from a single nodeSimple flooding and probabilistic flooding
Spread messages in short time, but cannot decrease messages
Proposed methodDelivers to 85% of nodes in short timeDecreases about 50% of messages
Periodic exchange delivers information to all nodes
Evaluation result of single flooding in 1000-node power-law network
Num
ber o
f nod
es
rece
ived
info
rmat
ion
(cum
ulat
ive)
Num
ber o
f con
trol
mes
sage
s (c
umul
ativ
e)
Time (sec)
Full Flooding
SimpleProbabilityMethod
FSLS
Proposed Method
0
1000
2000
3000
4000
5000
6000
7000
8000
0 1 2 3 4 5 6
Simple Flooding
Probabilistic flooding
FSLS
Proposed Method
0
200
400
600
800
1000
0 1 2 3 4 5 6Probabilistic flooding Probabilistic flooding
decreases 10% of decreases 10% of messagesmessages
Proposed method Proposed method decreases 50% of decreases 50% of
messagesmessages
Probabilistic flooding Probabilistic flooding spreads information in spreads information in short timeshort time
Periodical exchange Periodical exchange makes the nodes which makes the nodes which didndidn’’t receive messages t receive messages
to be informedto be informed
Proposed method Proposed method delivers information delivers information
to 85% of nodes to 85% of nodes quicklyquickly
3
1313Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka UniversityEvaluation of proposed method (3/3)
Flooding from different nodes
Flooding frequencyFlooding performed based on Poisson arrival with a mean 1 / sec
Proposed methodReduced 50% of messages compared to simple flooding
Num
ber o
f con
trol
mes
sage
s (c
umul
ativ
e)N
umbe
r of n
odes
rece
ived
in
form
atio
n (c
umul
ativ
e)
Time (sec)0
100000200000300000400000500000600000700000
800000
0 5 10 15 20 25 30 35 40 45 50
Simple flooding
SimpleProbabilityMethod
FSLS
Proposed Method
0
10000
20000
30000
40000
50000
60000
70000
0 5 10 15 20 25 30 35 40 45 50
FSLS
Proposed Method
0
10000
20000
30000
40000
50000
60000
70000
0 5 10 15 20 25 30 35 40 45 50
Simple flooding
Probabilistic flooding
Evaluation result of flooding from different nodes in 1000-node power-law network
Reduced 50% Reduced 50% of messages of messages compared to compared to
simple floodingsimple flooding
Reduced 40% Reduced 40% of messages of messages compared to compared to probabilistic probabilistic
floodingflooding
Our method scales better than simple flooding
1414Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University
Conclusion and Future worksA flooding in power-law network
Didn’t scale well due to the concentration of messages at a part of nodes
Proposed flooding methodProbabilistic flooding with periodic exchange of informationDecreased 50 % of control messages compared to simple flooding method
Future worksTheoretical deriving of relaying probabilityEvaluate in other network models that generate power-law attribute
1515Oct 05, 2005Oct 05, 2005 APCC2005APCC2005
Advanced Network Architecture Research GroupAdvanced Network Architecture Research Grouphttp://http://www.anarg.jpwww.anarg.jp//
Osaka UniversityOsaka University