Tuning EASY-Backfilling Queues

Abstract

Abstract. EASY-Backfilling is a popular scheduling heuristic for allocating jobs in large scale High Performance Computing platforms. While its aggressive reservation mechanism is fast and prevents job starvation, it does not try to optimize any scheduling objective per se. We consider in this work the problem of tuning EASY using queue reordering policies. More precisely, we propose to tune the reordering using a simulationbased methodology. For a given system, we choose the policy in order to minimize the average waiting time. This methodology departs from the First-Come, First-Serve rule and introduces a risk on the maximum values of the waiting time, which we control using a queue thresholding mechanism. This new approach is evaluated through a comprehensive experimental campaign on five production logs. In particular, we show that the behavior of the systems under study is stable enough to learn a heuristic that generalizes in a train/test fashion. Indeed, the average waiting time can be reduced consistently (between 11% to 42% for the logs used) compared to EASY, with almost no increase in maximum waiting times. This work departs from previous learning-based approaches and shows that scheduling heuristics for HPC can be learned directly in a policy space.

One Can Only Gain by Replacing EASY Backfilling: A Simple Scheduling Policies Case Study

“…However, such jobs are expected to be very few. This explains some results found by Lelong et al [21], in which they state that the SQF policy did not lead to many backfilling decisions in their experiments.…”

Adapting Batch Scheduling to Workload Characteristics: What Can We Expect From Online Learning?

“…All the comparisons and the techniques in the remainder of this paper are applied to the primary queue and the backfilling policy to fixed to SPF. We chose this setting because in [2], Lelong et al showed that reordering the primary queue is more beneficial than simply reordering the backfilling queue and in [6], The authors showed that SPF is a good policy for backfilling.…”

“…However, these works do not address dependencies between the workload and the objectives [10]. The dynP scheduler [2] offers an online approach to tune EASY queue. dynP requires the full simulated schedule for each of the candidate policies in every scheduling step, which makes the scheduling cost much higher than simple EASY.…”

“…A high threshold grants the scheduling policy a lot of freedom but low priority jobs risk starvation. In this work, the threshold is fixed at 2.10 5 seconds which roughly translates to 2.31 days [2].…”

AccaSim: a customizable workload management simulator for job dispatching research in HPC systems

“…However, any customization to the simulator needs to be done at the source code level, which can be complicated and error-prone. PYSS [23,21,28] and OCS [15] have similar characteristics to Alea, but provide less advanced WMS features as they are developed primarily for a specific research work in dispatching. In general, simple simulators like PYSS and OCS hinder the design of novel advanced dispatchers and their evaluation which requires a more flexible way to represent a WMS.…”