Reliability-Aware Speedup Models for Parallel Applications with Coordinated Checkpointing/Restart

摘要: Speedup models are powerful analytical tools for evaluating and predicting the performance of parallel applications. Unfortunately, well-known speedup like Amdahl’s law Gustafson’s do not take reliability into consideration therefore cannot accurately account application in presence failures. In this study, we enhance by considering impact failures effect coordinated checkpointing/restart. Unlike existing studies relying on Exponential failure distribution alone, work consider both Weibull distributions construction our reliability-aware models. The derived validated through trace-based simulations under a variety parameter settings. Our demonstrate these can effectively quantify speedup. Moreover, present two case to illustrate use