Distributed Strategies for Computational Sprints.

Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe a sprinting architecture in which many, independent chip multiprocessors share a power supply and sprints are constrained by the chips’ thermal limits and the rack’s power limits. More...

Descripción completa

Detalles Bibliográficos
Publicado en:Communications of the ACM Vol. 62; no. 2; pp. 98 - 107
Autores principales: Songchun Fan, Zahedi, Seyed Majid, Lee, Benjamin C.
Formato: Artículo
Publicado: Association for Computing Machinery Feb2019
Materias:
Acceso en línea:Ver este registro en EBSCOhost
fields @attributes:
  recordID: 1
pdfLink:
plink: https://search.ebscohost.com/login.aspx?direct=true&db=hlh&AN=134383090&site=ehost-live
header:
  @attributes:
    shortDbName: hlh
    uiTerm: 134383090
    longDbName: Humanities International Complete
    uiTag: AN
  controlInfo:
    bkinfo:
    jinfo:
      jid:
        00010782
        ACM
      jtl: Communications of the ACM
      issn: 00010782
      maglogo: N
    pubinfo:
      dt: Feb2019
      vid: 62
      iid: 2
      pid: 68
      pub: Association for Computing Machinery
    artinfo:
      ui:
        134383090
        10.1145/3299885
      ppf: 98
      ppct: 9
      formats:
      tig:
        atl: Distributed Strategies for Computational Sprints.
      aug:
        au:
          Songchun Fan
          Zahedi, Seyed Majid
          Lee, Benjamin C.
        affil:
          Duke University, California, USA.
          Duke University, Durham, NC, USA.
      su:
        Computational mathematics
        Computer programming
        Systems design
        Technical specifications
        Systems development
      sug:
        subj:
          Computational mathematics
          Computer programming
          Systems design
          Technical specifications
          Systems development
      ab: Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe a sprinting architecture in which many, independent chip multiprocessors share a power supply and sprints are constrained by the chips’ thermal limits and the rack’s power limits. Moreover, we present the computational sprinting game, a multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based on application phases and system conditions. The game produces an equilibrium that improves task throughput for data analytics workloads by 4–6× over prior greedy heuristics and performs within 90% of an upper bound on throughput from a globally optimized policy.
      pubtype: Periodical
      doctype: Article
      src: R
    language: English
    refInfo:
    copyright:
      @attributes:
        flag: Y
      dt:
        @attributes:
          year: 2019
    holdings:
      @attributes:
        islocal: N