Towards Cloud Bursting for Extreme Scale Supercomputers (Unrefereed Workshop Manuscript)

• Item type: SIG Technical Reports(1)
• 公開日: 2014-07-21
• タイトル: Towards Cloud Bursting for Extreme Scale Supercomputers (Unrefereed Workshop Manuscript)
• タイトル / 言語: en
• タイトル: Towards Cloud Bursting for Extreme Scale Supercomputers (Unrefereed Workshop Manuscript)
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: クラウド・ビッグデータ処理
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Tokyo Institute of Technology
• 著者所属: Tokyo Institute of Technology
• 著者所属: Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Tokyo Institute of Technology
• 著者所属(英): en
• 著者所属(英): Tokyo Institute of Technology
• 著者名: Tianqi, Xu; Kento, Sato; Satoshi, Matsuoka
• 著者名(英): Tianqi, Xu; Kento, Sato; Satoshi, Matsuoka
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Extreme-scale HPC systems, which consist of a large number of compute nodes, can provide high computational capacity for multiple users. However, computing nodes in the systems occasionally can not meet the demand due to bursty job requests in short period times. In order to accommodate the bursty requests, we consider federating HPC systems with public clouds, which is known as cloud bursting. Although the federated systems can acquire virtually infinite computational power with cloud bursting, the QoS may not be guaranteed due to a significant performance gap between HPC systems and public clouds. The most critical problem is a gap in I/O performance. In this paper, we propose an I/O acceleration technique using distributed cloud bursting buffers. We also create the I/O performance model to explore the effectiveness. Our model-based simulations, which target the TSUBAME supercomputer for an HPC system, and AMAZON EC2 for a public cloud, show that the distributed cloud busting buffer can improve I/O throughput while reducing the cost.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Extreme-scale HPC systems, which consist of a large number of compute nodes, can provide high computational capacity for multiple users. However, computing nodes in the systems occasionally can not meet the demand due to bursty job requests in short period times. In order to accommodate the bursty requests, we consider federating HPC systems with public clouds, which is known as cloud bursting. Although the federated systems can acquire virtually infinite computational power with cloud bursting, the QoS may not be guaranteed due to a significant performance gap between HPC systems and public clouds. The most critical problem is a gap in I/O performance. In this paper, we propose an I/O acceleration technique using distributed cloud bursting buffers. We also create the I/O performance model to explore the effectiveness. Our model-based simulations, which target the TSUBAME supercomputer for an HPC system, and AMAZON EC2 for a public cloud, show that the distributed cloud busting buffer can improve I/O throughput while reducing the cost.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10463942
• 書誌情報: 研究報告ハイパフォーマンスコンピューティング（HPC）; 巻 2014-HPC-145, 号 5, p. 1-8, 発行日 2014-07-21
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会