exploration of adaptation space -linking with efforts in
TRANSCRIPT
Fuyuki Ishikawa Associate Professor
Digital Content and Media Sciences Research DivisionNational Institute of Informatics (NII)
Exploration of Adaptation Space
- Linking with Efforts in Service-Oriented Computing
Adaptation is typically “switching solutions”According to changes of the environment, in particular, changes of availability/effectiveness of the solutions
Sep 10 2013 2f-ishikawa @ EASSy 2013
(Generalized) Overview
Goals Solution B
Solution A
Solution C
1. Overview
Identification of Solution Space?(at the design level, when solutions are reusable)
Sep 10 2013 3f-ishikawa @ EASSy 2013
(Generalized) Overview
Goals Solution B
Solution A
Solution C
1. Overview
Selection of “Attractive” Adaptation Space? inside the potential solution space(at the design level, for efficiency)
Sep 10 2013 4f-ishikawa @ EASSy 2013
(Generalized) Overview
Goals Solution B
Solution A
Solution C
1. Overview
Is adaptability (success rate) always enough?
Sep 10 2013 5f-ishikawa @ EASSy 2013
(Generalized) Overview
Goals Solution(Strong) Goals No/Less Adaptabliity
1. Overview
Is adaptability (success rate) always enough?Or, should we make a decision with trade-off?
Sep 10 2013 6f-ishikawa @ EASSy 2013
(Generalized) Overview
GoalsSolution
Weaker Goals More Adaptabliity
Goals Solution(Strong) Goals No/Less Adaptabliity
1. Overview
With one of our efforts in the area of Service-Oriented Computing
One instantiation of self-adaptive systems
With F. Wagner, B. Kloepper, S. Honiden,Towards Robust Service Compositions in the Context of Functionally Diverse Services,WWW 2012
Sep 10 2013 7f-ishikawa @ EASSy 2013
Try to Link…1. Overview
Function (input/output/precondition/effect)Goals as the whole, and consistency of the mid-flow
Quality (various criteria) including success rateOptimize under priorities and global constraints
Sep 10 2013 8f-ishikawa @ EASSy 2013
General Service Composition
Output: UnivResearcherID
Input: ResearcherID
Internet of Services
Total Price: <= $10 / invocationTotal Exec. Time <= 3 sec.Reliability of the Whole: >=99%Avg. User Score: >=3.5<Priorities: 0.4, 0.2, 0.1, 0.3>
Search Research Papersfrom a University
Selection of a Service Set
Input: UniversityName Output: List of Papers
1. Overview
Workflow of Tasks
In summary, the general problem has been:May be solved repeatedly with updated information at runtime (even partially during execution)
“Solution” under attention: service selection for each of the tasks
Sep 10 2013 9f-ishikawa @ EASSy 2013
General Service Composition1. Overview
Goals Solution B
Solution A
Solution C
Hard goals: functional consistency, global quality constraints
Soft goals: quality criteria (prioritized)
Analyze alternative services (i.e., potential solution space) at design/deployment-timeDerive the “best” part to be used at runtime
Avoid overhead and miss of optimality in “greedily deriving one best solution, repeatedly” at runtimeNaturally accompanies analysis of adaptability
Sep 10 2013 10f-ishikawa @ EASSy 2013
Our Work on Adaptive Compositions
[WWW'12]
2. Our Work
Goals Solution B
Solution A
Solution C
Hard goals: functional consistency, global quality constraints
Soft goals: total success rate,and best/avg./worst cases for other quality criteria (prioritized)
1
2
Construct “graphs” of service functionsDescendants are alternatives
Less/weaker input/precondMore/stronger output/postcond
used toAnalyze adaptabilityConstruct a “loose” plan (an adaptation space)(Efficiently check matching between connected services)
Sep 10 2013 11f-ishikawa @ EASSy 2013
Our Work on Adaptive Compositions
…
SS SSSSSSSS
SS SSSS
SS
SS
SS
SS SS
Creditcard-> Manga
Card/Bank/…-> Manga
Creditcard-> Book
SS
SelectRun
Adapt
…
SSSS
SS
2. Our Work
[WWW'12]
Derive “an adaptation space to be deployed “(for quick adaptations at runtime)
In any case, functionally-consistent, and global constraints satisfied“(Near-)Optimal” for given prioritiesTotal success rateBest/avg./worst cases for other quality criteria
By a custom, scalable genetic algorithm for this setting
Outperforms other methods in quality/scalability (details omitted)
Sep 10 2013 12f-ishikawa @ EASSy 2013
Our Work on Adaptive Compositions
1144225533
2. Our Work
[WWW'12]
Hotel search functions (output compatibility)(extracted from top 100 pages of Google search)
Sep 10 2013 13f-ishikawa @ EASSy 2013
Appendix: Example
Top search results do not mean rich functions
Alternatives decrease with strong goals(e.g., check pet allowance)
2. Our Work
Negligible differences should be defined to have a simpler graph…
Efforts on adaptive service compositionsviewed as exploration of adaptation space
Ongoing discussions:1. @runtime2. Use weaker services
(human intervention?)
Application case studies: under FP7 EU-Japan Project(IoT/Crowd as-a-Service)
Sep 10 2013 14f-ishikawa @ EASSy 2013
Summary and Directions3. Ongoing Direction
GoalsSolution
Weaker Goals More Adaptabliity
Goals Solution(Strong) Goals No/Less Adaptabliity
http://clout-project.eu/
Thank you!
[email protected]://research.nii.ac.jp/~f-ishikawa/en/
Sep 10 2013 15f-ishikawa @ EASSy 2013