• DocumentCode
    3470215
  • Title

    Policy of file migration at server in cluster file system

  • Author

    Shu, Jiwu ; Wang, Bing ; Weimin Meng ; Deng, Yiyan

  • Author_Institution
    Comput. Sci. & Technol. Dept., Tsinghua Univ., Beijing, China
  • fYear
    2004
  • fDate
    19-22 April 2004
  • Firstpage
    691
  • Lastpage
    698
  • Abstract
    File migration is a key technology to improve the efficiency of cluster file systems. In this paper, we give a thermal model that shows the file access frequency at each node of a cluster. Based on this model, we introduce a file migration policy, which can balance the workload of clustering server nodes. It is used to decide the optimal place of a single file in the cluster environment under a heavy workload. And the results of the simulation comparison experiments show that using this policy can decrease the average file access response time of a 4, 8, or 16 node cluster at a rate of about 35.52%, 62.43% and 49.07%, respectively. Furthermore, if the nodes are a uniform number, migration on the application server end will lower the accelerated speed of the average response time, i.e., the heavier the total workload, the more efficient the policy. In general, this policy can efficiently reduce the average file access response time, give a good extendibility and enhance the performance of the cluster file system as a whole.
  • Keywords
    cache storage; grid computing; performance evaluation; resource allocation; workstation clusters; cluster file system; clustering server nodes; file access frequency; file migration; performance; thermal model; workload balancing; Acceleration; Computer science; Costs; Delay; File servers; File systems; Frequency; Queueing analysis; Statistics; Tiles;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Cluster Computing and the Grid, 2004. CCGrid 2004. IEEE International Symposium on
  • Print_ISBN
    0-7803-8430-X
  • Type

    conf

  • DOI
    10.1109/CCGrid.2004.1336700
  • Filename
    1336700