{"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00218981","sets":["1164:1867:10898:10972"]},"path":["10972"],"owner":"44499","recid":"218981","title":["DPDKのRun-to-Completionモデルを用いたL2分散計算環境の提案"],"pubdate":{"attribute_name":"公開日","attribute_value":"2022-07-20"},"_buckets":{"deposit":"e64b91b9-1d4e-4a3d-9bbd-b4e43c548daa"},"_deposit":{"id":"218981","pid":{"type":"depid","value":"218981","revision_id":0},"owners":[44499],"status":"published","created_by":44499},"item_title":"DPDKのRun-to-Completionモデルを用いたL2分散計算環境の提案","author_link":["570562","570564","570563"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"DPDKのRun-to-Completionモデルを用いたL2分散計算環境の提案"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"ユーザ空間","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2022-07-20","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"名古屋工業大学大学院"},{"subitem_text_value":"名古屋工業大学大学院"},{"subitem_text_value":"名古屋工業大学大学院"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Nagoya Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"Nagoya Institute of Technology","subitem_text_language":"en"},{"subitem_text_value":"Nagoya Institute of Technology","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/218981/files/IPSJ-OS22156007.pdf","label":"IPSJ-OS22156007.pdf"},"date":[{"dateType":"Available","dateValue":"2024-07-20"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-OS22156007.pdf","filesize":[{"value":"730.2 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"11"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"ed4a18db-1f53-45f1-a467-c08748d6b81c","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2022 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"山本, 竜也"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"川島, 龍太"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"松尾, 啓志"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10444176","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8795","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"分散計算環境で実行する処理の中には，与えられたデータを繰り返し用いながら，小さいデータを計算機間でやりとりする処理がある．例えば，単回帰分析やロジスティック回帰などの機械学習である．これらの処理において，TCP/IP による制御は相対的に大きなオーバーヘッドとなる．また，カーネルによるパケット I/O 処理は DPDK のパケット I/O 処理に比べて低速である．本研究では，DPDK の Run-to-Completion モデルを用いた L2 分散計算環境を提案する．計算機間の通信に DPDK による L2 通信を用いることによって，TCP/IP による通信のオーバーヘッドを低減するとともに，カーネルによるパケット I/O 処理に比べて高速なパケット I/O 処理を用いることができる．また，Run-to-Completion モデルを採用することによって，Pipeline モデルを用いる場合に比べて CPU リソースや L1 キャッシュを有効活用できる．送られてきた 32 個の値を加算し送り返すプログラムを用いた事前実験 2 では，提案手法を用いた場合の処理性能は TCP/IP による通信を用いた場合に比べて，内部ループ回数が 1 回の場合は 30 倍，10 回の場合は 4 倍，100 回の場合は 0.4 倍高いことを確認した．また，単回帰分析を行うプログラムを用いた評価では，提案手法を用いた 2 台での分散処理の実行時間は，1 台での集中学習の場合に比べて 2 倍，TCP/IP による通信を用いた 2 台での分散処理の場合に比べて 1.2 倍高速であることを確認した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告システムソフトウェアとオペレーティング・システム（OS）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2022-07-20","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"7","bibliographicVolumeNumber":"2022-OS-156"}]},"relation_version_is_last":true,"weko_creator_id":"44499"},"id":218981,"updated":"2025-01-19T14:57:57.130013+00:00","links":{},"created":"2025-01-19T01:19:19.309219+00:00"}