@techreport{oai:ipsj.ixsq.nii.ac.jp:00211025, author = {村瀬, 香緒里 and 林, 真一 and 金子, 聡 and 村山, 耕一 and Kaori, Murase and Shinichi, Hayashi and Satoshi, Kaneko and Kouichi, Murayama}, issue = {3}, month = {May}, note = {データをオンプレミスに配置し,分析処理をパブリッククラウドで実行するデータ分析基盤は,データの詳細管理を可能とし,柔軟に処理性能を増減できる利点を併せ持つ.一方で,ハイブリッドクラウド構成では,オンプレミスとパブリッククラウドの設計方針の違いから,リソースの即時増減が難しいオンプレミスの性能に対 して分析処理の並列度を必要以上に高くするとコストパフォーマンスが下がる場合がある.本研究では,IT インフラに詳しくないデータ分析者でも前述の構成で適切な分析処理の並列度を決定できる方式を提案する.また本方式 によりコストパフォーマンスが低下しない適切な並列度を予測できる見込みを得た., A hybrid data analytics platform that consists of data lake on on-premise and servers for analytical processing on public cloud has the advantages of both detailed data management and flexible performance adjustment. However, for some cases cost performance declines due to design policy differences between on-premise and public cloud. For instance, cases where degree of parallelism for data analytics is set higher than necessary since it is difficult to change amount of processing resources of on-premise immediately. In this paper, we propose an optimization method for determining degree of parallelism based on allocatable on-premise resources of the hybrid data analytics platform. We prospect that it can predict appropriate degree of parallelism, and achieve an improved cost performance.}, title = {ハイブリッドクラウド構成における分析処理の並列度最適化方式}, year = {2021} }