Faculty of Computer Science Software Engineering Chair Is the PCM Ready for ACTORs and Multicore CPUs? A Use Case-based Evaluation Stefan Staude, Markus Frank, Marcus Hilbrich TU Chemnitz Faculty of Computer Science Software Engineering Chair SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude www.tu-chemnitz.de
Software-Performance-Prediction Prediction Software Engineer Model [Lehrig16] [Becker14] Single-Core [Frank16] Multi-Core SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 2/12 www.tu-chemnitz.de
Research Questions Q 1 : Is it possible to model a parallel system following the ACTORS approach with the PCM? Q 2 : How accurate are the simulated predictions compared to the real execution? SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 3/12 www.tu-chemnitz.de
Methodology Single-Core Multi-Core 3 1 2 4 Measurement Bank Transactions Evaluation 4 Step Implementation Palladio-Model SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 4/12 www.tu-chemnitz.de
Experimentsetup Bank Transactions Multiple Accounts with a positiv balance ● Transactions could fail and block each other ● Differing outcomes depending on the execution order ● Common use case for ACTORs ● Src. Acc. Acc. Actor Experiment Manager Accounts Transaction Trg. Acc. SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 5/12 www.tu-chemnitz.de
Palladio-Model Deployment Diagram: Src. Acc. Experiment Manager Trg. Acc. SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 6/12 www.tu-chemnitz.de
Palladio-Model SEFF ExperimentHandler: [...] SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 7/12 www.tu-chemnitz.de
Evaluation Reseach Question Q 1 : Each Actor must be modeled individually ● High manual effort ● No asynchrone message passing ● Our Model: Very abstract ● Loses ACTORs characteristics ● Low homomorphism ● SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 7/12 www.tu-chemnitz.de
Measurement Performance Results: Execution Simulation Worker Mean Time Speedup Mean Time Accuracy Threads 1 33.88 s 1.00 33.22 s 0.98 2 15.99 s 2.12 16.71 s 0.96 4 7.48 s 4.53 8.41 s 0.86 8 6.01 s 5.63 4.25 s 0.70 16 5.89 s 5.75 2.18 s 0.37 SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 9/12 www.tu-chemnitz.de
Measurement Performance Results: Execution Simulation Worker Mean Time Speedup Mean Time Accuracy Threads 1 33.88 s 1.00 33.22 s 0.98 2 15.99 s 2.12 16.71 s 0.96 4 7.48 s 4.53 8.41 s 0.86 8 6.01 s 5.63 4.25 s 0.70 16 5.89 s 5.75 2.18 s 0.37 SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 9/12 www.tu-chemnitz.de
Evaluation Reseach Question Q 2 : 40 Execution 35 Simulation 30 Mean Time in s 25 20 15 10 5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Threads SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 10/12 www.tu-chemnitz.de
Evaluation Reseach Question Q 2 : 18 Execution 16 Simulation 14 12 Speedup 10 8 6 4 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Threads SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 11/12 www.tu-chemnitz.de
Software-Performance-Prediction Prediction Software Engineer Model Questions? Questions? Multi-Core SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 12/12 www.tu-chemnitz.de
Evaluation Reseach Question Q 2 : Accuracy 1.2 1 0.8 0.6 Accuracy 0.4 0.2 0 1 2 4 8 16 Threads SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude www.tu-chemnitz.de
[Frank16] – M. Frank and M. Hilbrich „Performance Prediction for Multicore Environments – An Experiment Report“ [Lehrig16] – S. Lehrig and S. Becker „Using Performance Models for Planning the Redeployment to Infrastructure-as-a-Service Environments: A Case Study“ [Becker14] – M. Becker, M. Platenius and S. Becker „Cloud Computing Reduces Uncertainties in Quality-of-Service Matching!“ SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude www.tu-chemnitz.de
Recommend
More recommend