Predicting DBMS Performance Across Distinct Hardware Environments

• Item type: SIG Technical Reports(1)
• 公開日: 2023-09-19
• タイトル: Predicting DBMS Performance Across Distinct Hardware Environments
• タイトル / 言語: en
• タイトル: Predicting DBMS Performance Across Distinct Hardware Environments
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 高性能計算
• 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: Keio University
• 著者所属: Yahoo Japan Corporation
• 著者所属: Yahoo Japan Corporation
• 著者所属: Yahoo Japan Corporation
• 著者所属: Yahoo Japan Corporation
• 著者所属: Cybozu Labs, Inc.
• 著者所属: Keio University
• 著者所属(英): en
• 著者所属(英): Keio University
• 著者所属(英): en
• 著者所属(英): Yahoo Japan Corporation
• 著者所属(英): en
• 著者所属(英): Yahoo Japan Corporation
• 著者所属(英): en
• 著者所属(英): Yahoo Japan Corporation
• 著者所属(英): en
• 著者所属(英): Yahoo Japan Corporation
• 著者所属(英): en
• 著者所属(英): Cybozu Labs, Inc.
• 著者所属(英): en
• 著者所属(英): Keio University
• 著者名: Tatsuhiro, Nakamori; Shohei, Matsuura; Takashi, Miyazaki; Sho, Nakazono; Taiki, Sato; Takashi, Hoshino; Hideyuki, Kawashima
• 著者名(英): Tatsuhiro, Nakamori; Shohei, Matsuura; Takashi, Miyazaki; Sho, Nakazono; Taiki, Sato; Takashi, Hoshino; Hideyuki, Kawashima
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: DBMS performance heavily relies on tunable configuration knobs, and constructing a performance model to predict performance for certain configuration is crucial. While recent automatic database tuning approaches have focused on constructing a performance model specific to one production environment, there has been limited research conducted on transferring the knowledge of performance models across different hardware environments. Previous works have suggested that linear models are effective in transferring the performance models of binary configuration knobs. However, no research has been conducted regarding the transfer of performance models of continuous configuration knobs. Given that many influential tuning knobs in databases are continuous, assessing the viability of the linear models for continuous knobs becomes essential to ensure the effective transfer of performance models in practical databases. We investigate the feasibility of using linear models to transfer performance models of DBMS from one hardware environment to another. Our evaluation shows that linear models have the capability to predict target environment performance accurately in some cases. However, there were also several cases in which linear models failed to transfer performance models effectively. For linear models to work effectively, we need to be aware of the influence of hardware limitations on the effect of configuration knobs and identify configuration knobs that do not affect performance.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: DBMS performance heavily relies on tunable configuration knobs, and constructing a performance model to predict performance for certain configuration is crucial. While recent automatic database tuning approaches have focused on constructing a performance model specific to one production environment, there has been limited research conducted on transferring the knowledge of performance models across different hardware environments. Previous works have suggested that linear models are effective in transferring the performance models of binary configuration knobs. However, no research has been conducted regarding the transfer of performance models of continuous configuration knobs. Given that many influential tuning knobs in databases are continuous, assessing the viability of the linear models for continuous knobs becomes essential to ensure the effective transfer of performance models in practical databases. We investigate the feasibility of using linear models to transfer performance models of DBMS from one hardware environment to another. Our evaluation shows that linear models have the capability to predict target environment performance accurately in some cases. However, there were also several cases in which linear models failed to transfer performance models effectively. For linear models to work effectively, we need to be aware of the influence of hardware limitations on the effect of configuration knobs and identify configuration knobs that do not affect performance.
• 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 2023-HPC-191, 号 6, p. 1-10, 発行日 2023-09-19
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8841
• 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