Making Speculative Scheduling Robust to Incomplete Data

Abstract

In this work, we study the robustness of Speculative Scheduling to data incompleteness. Speculative scheduling has allowed to incorporate future types of applications into the design of HPC schedulers, specifically applications whose runtime is not perfectly known but can be modeled with probability distributions. Preliminary studies show the importance of speculative scheduling in dealing with stochastic applications when the application runtime model is completely known. In this work we show how one can extract enough information even from incomplete behavioral data for a given HPC applications so that speculative scheduling still performs well. Specifically, we show that for synthetic runtimes who follow usual probability distributions such as truncated normal or exponential, we can extract enough data from as little as 10 previous runs, to be within 5% of the solution which has exact information. For real traces of applications, the performance with 10 data points varies with the applications (within 20% of the full-knowledge solution), but converges fast (5% with 100 previous samples). Finally a side effect of this study is to show the importance of the theoretical results obtained on continuous probability distributions for speculative scheduling. Indeed, we observe that the solutions for such distributions are more robust to incomplete data than the solutions for discrete distributions.

Optimization Metrics for the Evaluation of Batch Schedulers in HPC

“…It is a particularly important objective for an HPC platform that costs multiple-million of dollars yearly to operate. This is the main objective studied in [10][11][12].…”

Reservation and Checkpointing Strategies for Stochastic Jobs

“…In this section, we provide an approximation algorithm of the optimal strategy for continuous distributions with bounded support The result for continuous distribution is particularly important: we have shown in recent work [10] that continuous distributions gave strategies that allowed using small data samples to find an efficient strategy. Here, it returns an arbitrarily good quality solution with low complexity.…”