Upload
cs-ncstate
View
368
Download
0
Tags:
Embed Size (px)
Citation preview
IV&V FacilityResearch Heaven,
West Virginia
1
SA @ WV(software assurance
research at West Virginia)
Kenneth McGill NASA IV&V Facility Research [email protected]
Dr. Tim Menzies Ph.D. (WVU)Software Engineering Research Chairtim@menzies,com
IV&V FacilityResearch Heaven,
West Virginia
2
Why, what is software assurance?
• Definition:– Planned and systematic set of
activities – Ensures that software
processes and products conform to requirements, standards, and procedures.
• Goals:– Confidence that SW will do what is
needed when it’s needed.
Before bad software After bad software
• Why software assurance?–bad software can kill good hardware. –E.g. ARIANE 5: (and many others)
•Software errors in inertial reference system•Floating point conversion overflow
Ariane 5
IV&V FacilityResearch Heaven,
West Virginia
3
OSMA Software Assurance Research Program
• Office of Safety & Mission Assurance (Code Q- OSMA)• Five million per year• Applied software assurance research• Focus:
– Software, not hardware– SW Assurance– NASA-wide applicability
• Externally valid results; i.e. useful for MANY projects
• Organization:– Managed from IV&V Facility– Delegated Program Manager: Dr. Linda Rosenberg, GSFC
IV&V FacilityResearch Heaven,
West Virginia
4
Many projects
• Mega: highest-level perspective– e.g. project planning tools like ASK-PETE
[Kurtz]• Macro:
– e.g. understanding faults [Sigal, Lutz & Mikulski]
• Micro:– e.g. source code browsing [Suder]
• Applied to basic:– Applied:
• (e.g.) MATT/RATT [Henry]: support large scale runs of MATLAB
– Basic (not many of these)• e.g. Fractal analysis of time series data
[Shereshevsky]• Many, many more
– Too numerous to list– Samples follow– See rest of SAS!
Horn of plenty
IV&V FacilityResearch Heaven,
West Virginia
5
Many more projects!
0
7
11 12
6 5
1 13
1
6
2
7
27
1012
4
0 0
5
26
22
0
5
10
15
20
25
30
AR
C
GR
C
GS
FC
IV&
V
JP
L
JS
C
KS
C
La
RC
MS
FC
Ind
us
try
Un
ive
rsit
y
2002
2003Total proposals: 2.2NASA centers: 1.5Industry: 26University: 3.7
Ratio FY02/FY01
Good news!• More good proposals than we can fund
Bad news!• same as the good news
IV&V FacilityResearch Heaven,
West Virginia
6
A survey of 44 FY01 CSIPs
project 1 2 3 4 5 6 7 8 9 10 11 12 13 14 to 44
AATT 2
ISS 2
Space Shuttle 2
ST5 2
Aura 1
CHIPS 1
CLCS 1
CM2 1
CMMI 1
DSMS 1
EOSDIS 1
FAMS 1
GLAST 1
HSM4 1
HST 1
Mars 07 1
Mars 08 1
PCS 1
Space Station 1
Starlight 1
Stereo 1
SWIFT 1
X-38 1
5 4 3 2 2 2 2 2 1 1 1 1 1 0
Need moretransitions!
(but don’tforget thetheory)
75% with noclaim forprojectconnections
IV&V FacilityResearch Heaven,
West Virginia
7
Action plan- restructure CSIPS: more transitions!
• New (year 1)– Fund many
• Renewed (year 2)– Continue funding the promising new
projects– Recommended: letter of endorsement
from NASA project manager• Transition (year 3)
– Select a few projects– Aim: tools in the hands of project folks– Required: project manager involvement
• Reality check:– Transition needs time– Data drought
IV&V FacilityResearch Heaven,
West Virginia
8
Long transition cycles
CO2 + 2H2 —> CH4 + O2
Marsatmosphere
oxidizerfuel
on-board
(no photo)
CarmenMikulski
JPL
Robyn LutzJPL, CS-Iowa
State
• Pecheur & practical formal methods
– In-Situ Propellant Production project– Taught developers:
• Livingstone model-based diagnosis
• model-checking tool tools• developed by Reid Simmons,
(CMU)– Technology to be applied to the
Intelligent Vehicle Health Maintenance (IVMS) for 2nd generation shuttles
• Lutz, Mikulski & ODC-based analysis of defects
– Deep-space NASA missions– Found 8 clusters of recurring defects– Proposed and validated 5
explanations of the clusters– Explanations changes to NASA
practices– ODC being evaluated by JPL’s defect
management tool team
Charles Pecheur
RIACS, ASE, ARC
IV&V FacilityResearch Heaven,
West Virginia
10
End the drought:bootstrap off other systems
• Find the enterprise-wide management information system
• Insert data collection hooks– E.g. JPL adding
ODC to their defect tracking system
– WVU SIAT sanitizer
IV&V FacilityResearch Heaven,
West Virginia
11
End the drought:Contractors as researchers
active data repository
• Buy N licenses of a defect tracking tool (e.g. Clearquest)
• Give away to projects– In exchange for their data
• Build and maintain a central repository for that data– With a web-based query
interface
• Data for all
take me to your data
IV&V FacilityResearch Heaven,
West Virginia
12
End the drought:Contractors as researchers (2)
abstractionabstractionactionaction reflectionreflection
experienceexperience 1
2
3
4
Mark SuderTitan, IV&V
Hypertext power browser for source code4 SIAT-1}
For high-severity errors, recall what SIAT querieslead to finding those errors
4’
2’
Assess each such “power queries”Reject the less useful ones
3’
Procedures manual for super SIAT ornew search options in interface
SIAT2}1’ Use it.
See also:
• Titan’s new ROI project
• Any contractor proposing an NRA
• Galaxy Global’s metric project
See also:
• Titan’s new ROI project
• Any contractor proposing an NRA
• Galaxy Global’s metric project
IV&V FacilityResearch Heaven,
West Virginia
13
End the drought:raid old/existing projects
• Cancelled projects with public-domain software– E.g. X-34
• Or other open source NASA projects – E.g. GSFC’s ITOS:– real-time control and
monitoring system during development, test, and on-orbit operations,
– UNIX, Solaris, FreeBSD, Linux, PC
– Free!!– NASA project connections:
• Triana, • Swift, • HESSI, • ULDB, • SMEX, • Formation Flying Testbed, • Spartan
IV&V FacilityResearch Heaven,
West Virginia
14
End the drought:synergy groups
• N researchers– Same task– Different
technologies• Share found data• E.g. IV&V business
case workers• E.g. monthly fault
teleconferences– JPL:
• Lutz, Nikora
– Uni. Kentucky: • Hayes
– Uni. Maryland: • Smidts
– WV: • Chapman
(Galaxy Global) & Menzies (WVU)
IV&V FacilityResearch Heaven,
West Virginia
15
End the drought:Tandem experiments
• “Technique X finds errors”– So?
• Industrial defect detection capability rates: – TR(min,mean,max)– TR(0.35, 0.50, 0.65)– Assumes manual
“Fagan inspections”• Is “X” better than a
manual 1976technique?
• Need “tandemexperiments”to check
• I.e. do it twice– Once by the researchers– Once by IV&V
contractors (baseline)
0
20
40
60
80
100
120
defects found
analysis design code test
baseline FM Fagan
fictional data
0
20
40
60
80
100
120
cost
analysis design code test
IV&V FacilityResearch Heaven,
West Virginia
16
Alternatively:End your own drought
• Our duty, our goal:– Work the data problem (e.g. see above)– Goal of CI project year1: build bridges– But the more workers, the better
• Myth: there is a “data truck” parked at IV&V – full of goodies, just for you
• Reality: Access negotiation takes time– With contractors, within NASA
• We actively assist:– Each connection is a joy to behold,
an occasion to celebration– We don’t celebrate much
• Bottom line:– We chase data for dozens of projects– Researchers have more time, more focus on
their particular data needs• Ken’s law:
– $$$ chases researchers who chase projects– CI year2, year3: needs a project connection
IV&V FacilityResearch Heaven,
West Virginia
17
Alternatively (2), accept the drought and sieve the dust
• The DUST project:– Assumes a few key options control the rest
• Methodology:– Simulate across range of options
– Data dust clouds– Too many options: what leads to what?
– Summarize via machine learning– Condense dust cloud– Improve mean, reduce variance
• Case studies: – JPL requirements engineering:
• Feather/JPL [Re02]
– Project planning: • DART- Raque/ IVV; Chaing/UBC; • IV&V costing: Marinaro/IVV, Smith/WVU• general: Raffo, et.al/PSU [Ase02]
– An analysis of pair programming: Smith/WVU– Better predictors for:
• testability: Cukic/WVU, Owen/WVU [Issre02, Ase02]• faults: diStefano/WVU, McGill/IVV; Chapman/GG• reuse : diStefano/WVU [ToolsWithAI02]
Figure 2. Initial (scattered black points) and Final (dense white points)
0
50
100
150
200
250
300
0 300000 600000 900000 1200000Cost
Ben
efit
Each dot = 1 random
project plan
The answer my friend, is blowin’
in the wind
But wait: the times they
are changing
IV&V FacilityResearch Heaven,
West Virginia
18
Katerina Goseva Popstojanova
Other WVU SA research
Architecturaldescriptions
Fault, failure
data oncomponents,
connectors
SoftwareSpecs & design(early life cycle)
Code analysis(iv&v,operational usage)
Metrics(complexity,coupling,entropy )
Failure data from testing
Severity of failures
UML (sequence diagrams, state charts)
UML simulations
Static (SIAT, Mccabe, entrophy)
Dynamic (testing, runtime monitoring)
Testing & formal methods
Bayesian approach to reliability
Architectural metrics
Risk assessment & dynamic UML
Reliability & operational profile errors
Hany Ammar
Bojan Cukic
collaborator
Goal: accurate, stable, risk assessment early in the
lifecycle
Goal: accurate, stable, risk assessment early in the
lifecycle
IV&V FacilityResearch Heaven,
West Virginia
19
More WVU research (FY02 UIs)
Architectural metrics
Risk assessment & dynamic UML
Intelligent flight controllers
Testing & formal methods
Bayesian approach to reliability
Fractal study of resource dynamics
Reliability & operational profile errors
SE research chair
interns
DUST
Ammar
Cukic
Goseva-Popstojanova
Menzies
newrenewed
c = conference w = workshopj = journal
ISS hub controller, “Dryden application”
F15
“JPL deep space mission”DART“KC-2”IVV cost models
SIAT
X34
ITOS
X38
jj
j, ccccccc, w
c
cccccc
jc
c
w
FY03 proposals = 2.2*FY02
IV&V FacilityResearch Heaven,
West Virginia
20
Function Point Metrics for Safety-Critical Software
• Thesis:– Traditional function-point
cost estimation– Incorrect for safety-critical
software
• > 1 way to skin a cat– >1 way to realize a safety
critical function:– NCP=
N-copy programming– NVP=
N-Version Programming ,
– NSCP= N Self-Checking Programming,
– …– With, without redundancy,
• Method: – explore them all!
1.3000
1.4000
1.5000
1.6000
1.7000
1.8000
1.9000
2.0000
0 0.033 0.1 0.33 1
Algorithm Complexity
H2/H
1, C
2/C1
NCP
NVP, NSCP
RFCS
CRB
RB, NRB
DRB, EDRB
NCP
NVP, NSCP
RFCS
CRB
RB, NRB
DRB, EDRB
Design Diversity, add eight more
Design Diversity, add onemore
Data Diversity
H2 and C2 : effort & cost, redundant systemH1 and C1: effort & cost, non-redundant system Afzel Noore
IV&V FacilityResearch Heaven,
West Virginia
21
Pre-disaster warnings [Cukic, Shereshevsky]
Can we defer a maintenance cycle and keep doing science for a while longer?
Mark Shereshevsky
CrashEarly warning
}
Time for gracefulshutdown
Bojan Cukic
ARTS II
IV&V FacilityResearch Heaven,
West Virginia
22
Intelligent flight controllers [Napolitano, Cukic] (and menzies)
Marcello Napolitano(Mechanical and
Aerospace)
Bojan Cukic(CSEE)
Lifecycle opportunities for V&V of neural network based adaptive control systems.