Effective Processing of Metadata Operations in Large- Scale Distributed File Systems

When metadata is distributed across many metadata servers in a distributed file system with a locality-preserving metadata distribution unit like a directory, simply increasing the number of metadata servers does not give high-aggregate performance for the metadata operation such as file creation. This paper proposes an effective and consistent method for processing metadata operations, at low cost, without depending on the traditional 2 phase commit protocol. Designing an alternate protocol, we showed that the performance limitations of metadata operations, caused by the distributed environment, can be overcome and the materialization of a large-scale distributed file system is actually possible.