Per Opteryx Automatically Improve Software Architecture Models for

Per. Opteryx Automatically Improve Software Architecture Models for Performance, Reliability, and Costs using Evolutionary Algorithms Anne Martens Karlsruhe Institute of Technology (KIT), Germany Heiko Koziolek, Steffen Becker, Ralf Reussner WOSP / SIPEW 2010 www. kit. edu

Software Performance Engineering C 1 3 sec A Transform Solve Change component and deployment 2. 5 sec C 2 A Transform Solve Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 2 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Not only Performance! C 1 A Transform Solve 3 sec p(fail) 0. 01% $ 5700 Solve 2. 5 sec p(fail) 0. 02% $ 12000 Change component and deployment C 2 A Transform Optimise multiple criteria at once Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 3 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Multicriteria Optimisation Architectural Candidate Time A 5 s $40 K Costs Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 4 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Multicriteria Optimisation Generated & Evaluated Time A 5 s $40 K Costs Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 5 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Multicriteria Optimisation Time A 5 s B 3 s C 2 s Pareto-optimal $20 K $33 K $40 K Costs Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 6 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Related Work: Quality Optimisation Rule-based approaches: Single quality only Parsons 2008, Cortellessa 2009, Performance. Booster (Xu&Woodside 2008), Arch. E (Mc. Gregor 2007) Multicriteria evaluation: No improvement Bondarev 2007, Grunske 2007 Optimisation: Limited degrees of freedom Arche. Opteryx (Aleti 2009), Canfora 2005, Kavimandan 2009, Sassy (Menascé 2010) Missing: Flexible multicriteria optimisation at the design level Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 7 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Per. Opteryx Approach Flexible degrees of freedom Multiple qualities Multi-criteria optimization Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 8 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Degrees of Freedom Design decision that can still be made C Variation point Which instance to use for component type C? Range of options C 1, C 2, or C 3 Degree of freedom Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 9 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Types of Degrees of Freedom in CBSE Software Component selection Middleware selection Component replication Software configuration Deployment Allocation Processing Rate Number of Servers Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 10 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Instances of Degrees of Freedom Degree Matching Rule Allocation Each component Processor speed Each server Component selection Search alternatives . . . Component selection for C Processor speed of server 1 C D Component selection for D Allocation of C Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 11 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Search Problem Degree Choice evaluate Component selection C C 2 Allocation C Server 1 Speed server 1 2 GHz . . . Response in 2. 5 s P(failure) 0. 02% Cost $6000 transform evaluate initial model candidate model Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 12 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Search Implementation NSGA-II [Deb 2002] Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 13 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Quality evaluation Palladio Component Model [Becker 2007] PCM 2 LQN PCM 2 DTMC PCM 2 Cost [Koziolek 2008] [Brosch 2009] [Martens 2010] Performance Reliability Cost Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 14 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Case Study with Per. Opteryx (1/2) Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 15 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Case Study with Per. Opteryx (1/2) 1235 candidates 58 Pareto optimal 8 h running time Response Time Component allocation Processing rates Component selection OD F O Cost s P Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 16 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Average Response Time (Seconds) Case Study with Per. Opteryx (2/2) 20 All candidates 18 Pareto-optimal candidates 16 Initial Candidate 14 12 10 8 6 4 RT: 1. 34 s POFOD: 5. 2 E-4 Cost: 69. 83 Only four, but faster servers Different Webserver RT: 2. 2 s POFOD: 6 E-4 Cost: 98 2 0 0 20 40 60 80 100 Costs (K$) 120 140 160 Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 17 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Future Work Short term • Performance heuristics • Requirement support • More degrees of freedom Long term • Handle uncertainty of predictions • Qo. S process integration ? Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 18 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD

Conclusions Automated Architecture Improvement Flexible degrees of freedom Multiple qualities Multi-criteria Optimization http: //sdqweb. ipd. kit. edu/wiki/Per. Opteryx Motivation – Related Work – Approach – Case Study – Future Work –Conclusion 19 13. 06. 2021 Anne Martens Software Design and Quality Group, IPD