A Zones-Based Metadata Management Method for Distributed File System

It is challengeable for the single server to manage the large metadata in MooseFS which is full of small files. This paper proposes a zones-based metadata management (ZMM) method for distributed file system. ZMM changes a single metadata server into a cluster, which includes a zone server and several sub-metadata servers. We tested ZMM in the platform named SandBox and the results shown that the throughput of writing and reading metadata was improved by using ZMM compared with using the single metadata server in MooseFS. With the increase of sub-metadata servers, the throughput of writing and reading metadata was improved as well.

[1]  Fang Wang,et al.  MHS: A distributed metadata management strategy , 2009, J. Syst. Softw..

[2]  Alan L. Cox,et al.  The Hadoop distributed filesystem: Balancing portability and performance , 2010, 2010 IEEE International Symposium on Performance Analysis of Systems & Software (ISPASS).

[3]  Mahadev Satyanarayanan,et al.  Andrew: a distributed personal computing environment , 1986, CACM.

[4]  Samuel Lang,et al.  GIGA+: scalable directories for shared file systems , 2007, PDSW '07.

[5]  Howard Gobioff,et al.  The Google file system , 2003, SOSP '03.