Font Size: a A A

Research On The Metadata Management Of Windows Client File System

Posted on:2008-10-23Degree:MasterType:Thesis
Country:ChinaCandidate:Q FuFull Text:PDF
GTID:2178360272969250Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
The object based storage system(OBS) which apply access to the object and use tripartite communications to overcome block and file interface defects , was introduced to construct a pera-byte level storage system. With the defined rich object interface and the empoldered surplus computing resources of the object based storage device, the OBS system is intelligent, platform-crossing and scalable. While the client is implemented on the Linux platform, the users whose application is on the Windows platform need to access the OBS system through Samba server, which lower the performance. And it is meaningful to implement a OBS client on the Windows platform.The metadata management whose operation rate is between 50% to 80% is a crucial component for the OBS system, while the characters of the OBS system such as long latency and frequent communication lead to the low network utilization. The client based metadata management overcomes these by applied active service to the user. The metadata is prefetched from the MDS as a batch and stored in the OBS client. The relevance of the metadata was established and the two-class metadata update approach is applied to reduce the latency and improve the network utilization. By the testing which uses the Geographic Information System's metadata usage as a trace, it is indicated that the average metadata latency is reduced 40% under the circumstance with the cache size 128.A correlation model for the metadata was introduced and with the analyzed parameter a correlation replacement algorithm was developed. Test results show related replacement algorithm outperforms traditional replacement algorithm, with the largest increase rate hit 9% compared with LRU replacement algorithm.
Keywords/Search Tags:object based storage system, metadata management, correlation based replacement
PDF Full Text Request
Related items