Improving the Scheduling of Parallel Applications using Accurate AIC-based Performance Models
Predictions based on analytical performance models can be used on efficient scheduling policies, in order to establish adequate resources for an optimal execution in terms of throughput and response time. However, it is a hard job developing accurate analytical models of parallel applications. In this paper, an accurate performance model of the HPL benchmark is obtained in a easy way by means of AIC-based model selection methods provided by the TIA framework. The performance of backfilling policy algorithms on schedulers using this AIC-based model is analyzed in the GridSim simulator and compared to the results obtained using the theoretical analytical model provided by the authors of the benchmark
keywords:
Publication: Congress
1624015014641
June 18, 2021
/research/publications/improving-the-scheduling-of-parallel-applications-using-accurate-aic-based-performance-models
Predictions based on analytical performance models can be used on efficient scheduling policies, in order to establish adequate resources for an optimal execution in terms of throughput and response time. However, it is a hard job developing accurate analytical models of parallel applications. In this paper, an accurate performance model of the HPL benchmark is obtained in a easy way by means of AIC-based model selection methods provided by the TIA framework. The performance of backfilling policy algorithms on schedulers using this AIC-based model is analyzed in the GridSim simulator and compared to the results obtained using the theoretical analytical model provided by the authors of the benchmark - D. R. Martínez, J. L. Albín, V. Blanco, T. F. Pena, J. C. Cabaleiro, F. F. Rivera
publications_en