Zobrazit v EDS

Shisha: Online Scheduling of CNN Pipelines on Heterogeneous Architectures

Uloženo v:
Podrobná bibliografie
Název: Shisha: Online Scheduling of CNN Pipelines on Heterogeneous Architectures
Autoři: Soomro, Pirah Noor, 1993, Abduljabbar, Mustafa, Castrillon, Jeronimo, Pericas, Miquel, 1979
Zdroj: 14th International Conference on Parallel Processing and Applied Mathematics, PPAM 2022, Gdansk, Poland Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). 13826 LNCS:249-262
Témata: Design space exploration, CNN parallel pipelines, Processing on heterogeneous computing units, Online tuning, Processing on chiplets
Popis: Many modern multicore processors integrate asymmetric core clusters. With the trend towards Multi-Chip-Modules (MCMs) and interposer-based packaging technologies, platforms will feature heterogeneity at the level of cores, memory subsystem and the interconnect. Due to their potential high memory throughput and energy efficient core modules, these platforms are prominent targets for emerging machine learning applications, such as Convolutional Neural Networks (CNNs). To exploit and adapt to the diversity of modern heterogeneous chips, CNNs need to be quickly optimized in terms of scheduling and workload distribution among computing resources. To address this we propose Shisha, an online approach to generate and schedule parallel CNN pipelines on heterogeneous MCM-based architectures. Shisha targets heterogeneity in compute performance and memory bandwidth and tunes the pipeline schedule through a fast online exploration technique. We compare Shisha with Simulated Annealing, Hill Climbing and Pipe-Search. On average, the convergence time is improved by ∼ 35 × in Shisha compared to other exploration algorithms. Despite the quick exploration, Shisha’s solution is often better than that of other heuristic exploration algorithms.
Přístupová URL adresa: https://research.chalmers.se/publication/536220
Databáze: SwePub
FullText Text:
  Availability: 0
CustomLinks:
  – Url: https://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=EBSCO&SrcAuth=EBSCO&DestApp=WOS&ServiceName=TransferToWoS&DestLinkType=GeneralSearchSummary&Func=Links&author=Soomro%20PN
    Name: ISI
    Category: fullText
    Text: Nájsť tento článok vo Web of Science
    Icon: https://imagesrvr.epnet.com/ls/20docs.gif
    MouseOverText: Nájsť tento článok vo Web of Science
Header DbId: edsswe
DbLabel: SwePub
An: edsswe.oai.research.chalmers.se.e7cee521.ad27.4991.8cdc.fedcb0abc0c6
RelevancyScore: 856
AccessLevel: 6
PubType: Conference
PubTypeId: conference
PreciseRelevancyScore: 855.779541015625
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Shisha: Online Scheduling of CNN Pipelines on Heterogeneous Architectures
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Soomro%2C+Pirah+Noor%22">Soomro, Pirah Noor</searchLink>, 1993<br /><searchLink fieldCode="AR" term="%22Abduljabbar%2C+Mustafa%22">Abduljabbar, Mustafa</searchLink><br /><searchLink fieldCode="AR" term="%22Castrillon%2C+Jeronimo%22">Castrillon, Jeronimo</searchLink><br /><searchLink fieldCode="AR" term="%22Pericas%2C+Miquel%22">Pericas, Miquel</searchLink>, 1979
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <i>14th International Conference on Parallel Processing and Applied Mathematics, PPAM 2022, Gdansk, Poland Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)</i>. 13826 LNCS:249-262
– Name: Subject
  Label: Subject Terms
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Design+space+exploration%22">Design space exploration</searchLink><br /><searchLink fieldCode="DE" term="%22CNN+parallel+pipelines%22">CNN parallel pipelines</searchLink><br /><searchLink fieldCode="DE" term="%22Processing+on+heterogeneous+computing+units%22">Processing on heterogeneous computing units</searchLink><br /><searchLink fieldCode="DE" term="%22Online+tuning%22">Online tuning</searchLink><br /><searchLink fieldCode="DE" term="%22Processing+on+chiplets%22">Processing on chiplets</searchLink>
– Name: Abstract
  Label: Description
  Group: Ab
  Data: Many modern multicore processors integrate asymmetric core clusters. With the trend towards Multi-Chip-Modules (MCMs) and interposer-based packaging technologies, platforms will feature heterogeneity at the level of cores, memory subsystem and the interconnect. Due to their potential high memory throughput and energy efficient core modules, these platforms are prominent targets for emerging machine learning applications, such as Convolutional Neural Networks (CNNs). To exploit and adapt to the diversity of modern heterogeneous chips, CNNs need to be quickly optimized in terms of scheduling and workload distribution among computing resources. To address this we propose Shisha, an online approach to generate and schedule parallel CNN pipelines on heterogeneous MCM-based architectures. Shisha targets heterogeneity in compute performance and memory bandwidth and tunes the pipeline schedule through a fast online exploration technique. We compare Shisha with Simulated Annealing, Hill Climbing and Pipe-Search. On average, the convergence time is improved by ∼ 35 × in Shisha compared to other exploration algorithms. Despite the quick exploration, Shisha’s solution is often better than that of other heuristic exploration algorithms.
– Name: URL
  Label: Access URL
  Group: URL
  Data: <link linkTarget="URL" linkTerm="https://research.chalmers.se/publication/536220" linkWindow="_blank">https://research.chalmers.se/publication/536220</link>
PLink https://erproxy.cvtisr.sk/sfx/access?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsswe&AN=edsswe.oai.research.chalmers.se.e7cee521.ad27.4991.8cdc.fedcb0abc0c6
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1007/978-3-031-30442-2_19
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 14
        StartPage: 249
    Subjects:
      – SubjectFull: Design space exploration
        Type: general
      – SubjectFull: CNN parallel pipelines
        Type: general
      – SubjectFull: Processing on heterogeneous computing units
        Type: general
      – SubjectFull: Online tuning
        Type: general
      – SubjectFull: Processing on chiplets
        Type: general
    Titles:
      – TitleFull: Shisha: Online Scheduling of CNN Pipelines on Heterogeneous Architectures
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Soomro, Pirah Noor
      – PersonEntity:
          Name:
            NameFull: Abduljabbar, Mustafa
      – PersonEntity:
          Name:
            NameFull: Castrillon, Jeronimo
      – PersonEntity:
          Name:
            NameFull: Pericas, Miquel
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 01
              Type: published
              Y: 2023
          Identifiers:
            – Type: issn-print
              Value: 16113349
            – Type: issn-print
              Value: 03029743
            – Type: issn-locals
              Value: CTH_SWEPUB
          Numbering:
            – Type: volume
              Value: 13826 LNCS
          Titles:
            – TitleFull: 14th International Conference on Parallel Processing and Applied Mathematics, PPAM 2022, Gdansk, Poland Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
              Type: main
ResultId 1