Introducing Power-capping in Slurm scheduling

Abstract : The last decades have been characterized by an ever growing requirement in terms of computing and storage resources. This tendency has recently put the pressure on the ability to efficiently manage the power required to operate the huge amount of electrical components associated with state-of-the-art computing and data centers. The power consumption of a supercomputer needs to be adjusted based on varying power budget or electricity availabilities. As a consequence, Resource and Job Management Systems have to be adequately adapted in order to efficiently schedule jobs with optimized performance while limiting power usage whenever needed. Our goal is to introduce a new power consumption adaptive scheduling strategy that provides the capability to autonomously adapt the executed workload to the available or planned power budget. The originality of this approach relies on a combination of DVFS (Dynamic Voltage and Frequency Scaling) and node shut-down techniques.
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01102285
Contributor : Grégory Mounié <>
Submitted on : Monday, January 12, 2015 - 2:27:50 PM
Last modification on : Thursday, April 4, 2019 - 10:18:05 AM

Identifiers

  • HAL Id : hal-01102285, version 1

Collections

CEA | INRIA | UGA | DAM | LIG

Citation

David Glesser, Yiannis Georgiou, Matthieu Hautreux, Denis Trystram. Introducing Power-capping in Slurm scheduling. Slurm User Group Meeting 2014, 2014, Lugano, Switzerland. ⟨hal-01102285⟩

Share

Metrics

Record views

272