A Novel Metadata Management Architecture Based on Service Separation in Cluster File System

  • Authors:

  • Junwei Zhang;Jingliang Zhang;Jiangang Zhang;Xiaoming Han;Lu Xu

  • Affiliations:

  • -;-;-;-;-

  • Venue:
  • PDCAT '09 Proceedings of the 2009 International Conference on Parallel and Distributed Computing, Applications and Technologies
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Efficient metadata management is critical in large scale cluster file systems. Based on service separation idea, a novel metadata management architecture is presented in this paper to improve metadata access performance and scalability. Storage management is completely separated from namespace and data services. A storage management server cluster singly deals with storage management. To further reduce metadata server loads, parts of file attributes are also separated from metadata server to storage management server cluster. A prototype system has been designed and implemented, and tests have also been done. Measured performance results are very promising. The performance improvement is nearly linear as storage management server number increasing.