Evaluation of Parallel Data Restoration for Metro Area Distributed Storage System with Assumption of Storage Node Restart Trend under Large-Scale Disaster

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2021/03/01

タイトル(英語): Evaluation of Parallel Data Restoration for Metro Area Distributed Storage System with Assumption of Storage Node Restart Trend under Large-Scale Disaster

著者名: Hitoshi Kamei (Hitachi, Ltd.), Takaki Nakamura (Hitachi, Ltd.), Hiroaki Muraoka (Tohoku University)

著者名(英語): Hitoshi Kamei (Hitachi, Ltd.), Takaki Nakamura (Hitachi, Ltd.), Hiroaki Muraoka (Tohoku University)

キーワード: disaster recovery，parallel data restoration，distributed storage system

要約(英語): In this paper, the parallel data restoration is evaluated with assumption of the storage node restart trend under large-scale disaster. In our previous study, we evaluated a proposed method for the data replication and data restoration for large-scale disasters. The proposed data replication copies data among nearby storage nodes each other to ensure reachability. Additionally, to broaden throughput, the proposed data restoration gets the backup data in parallel. The erasure coding is applied to avoid data lost. In this way, the data of the broken storage nodes are restored as soon as possible. However, the evaluations of previous study do not cover the transitional situation of recovery. After large-scale disaster, it is supposed that a part of storage nodes destructs and some storage nodes widely stop due to blackout. After a while, the storage nodes that temporally stop restart gradually. Thus, the number of the working storage nodes gradually increases. Therefore, a challenge of the parallel data restoration is to examine the relation between the suppression of the load and the prolongation of the restoration execution time under the situation and reveal the appropriate configurations. In this study, the model and method for the evaluation of the parallel data restoration are created. Additionally, the load and the restoration execution time are evaluated. From the results, the configuration of two data chunks and 30 parity chunks is suitable when the throughput is high. Meanwhile, the configuration of eight data chunks and 24 parity chunks is suitable when the throughput is low and the gradient of the storage node restart trend is steep. Moreover, if the gradient of the storage node restart trend is gradual, then for each throughput, the PDR configuration with the smallest number of data chunks of the configurations that can restore backup data is suitable.

本誌: 電気学会論文誌C（電子・情報・システム部門誌） Vol.141 No.3 （2021） 特集：スマートシステムと計測・制御技術 －超スマート社会に向けて－

本誌掲載ページ: 483-493 p

原稿種別: 論文／英語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/141/3/141_483/_article/-char/ja/