Autoscaling of cores in multicore processors using power and thermal workload signatures

Rupesh Raj Karn, Ibrahim M. Elfadel

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Autoscaling of cloud clusters based on dynamic estimation of workloads is a well-known practice in data center management. However, this practice has not been widely adopted in the multicore processor area due to the lack of a real-Time workload classification front end. In this paper, we present a novel methodology for core autoscaling in multicore processors. The methodology is based on the identification of workload signatures at both the core and the thread level. In particular, correlations between thread-based performance counters are used to decide thread migration policies to maximize per-core utilization and reduce the number of active cores. Power, thermal, and performance-Aware autoscaling policies are presented, and extensive numerical experiments are used to illustrate the advantages of our algorithm for real-Time multicore power and performance management.

Original languageBritish English
Title of host publication2016 IEEE 59th International Midwest Symposium on Circuits and Systems, MWSCAS 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781509009169
DOIs
StatePublished - 2 Jul 2016
Event59th IEEE International Midwest Symposium on Circuits and Systems, MWSCAS 2016 - Abu Dhabi, United Arab Emirates
Duration: 16 Oct 201619 Oct 2016

Publication series

NameMidwest Symposium on Circuits and Systems
Volume0
ISSN (Print)1548-3746

Conference

Conference59th IEEE International Midwest Symposium on Circuits and Systems, MWSCAS 2016
Country/TerritoryUnited Arab Emirates
CityAbu Dhabi
Period16/10/1619/10/16

Keywords

  • Core
  • Correlation
  • Performance
  • Power
  • Signature
  • Temperature
  • Workload

Fingerprint

Dive into the research topics of 'Autoscaling of cores in multicore processors using power and thermal workload signatures'. Together they form a unique fingerprint.

Cite this