is the pcm ready for actors and multicore cpus a use case
play

Is the PCM Ready for ACTORs and Multicore CPUs? A Use Case-based - PowerPoint PPT Presentation

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


  1. 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

  2. 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

  3. 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

  4. 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

  5. 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

  6. Palladio-Model Deployment Diagram: Src. Acc. Experiment Manager Trg. Acc. SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 6/12 www.tu-chemnitz.de

  7. Palladio-Model SEFF ExperimentHandler: [...] SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 7/12 www.tu-chemnitz.de

  8. 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

  9. 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

  10. 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

  11. 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

  12. 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

  13. Software-Performance-Prediction Prediction Software Engineer Model Questions? Questions? Multi-Core SSP Karlsruhe ∙ 15.11.17 ∙ Stefan Staude 12/12 www.tu-chemnitz.de

  14. 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

  15. [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