A design of metadata server cluster in large distributed object-based storage

In large distributed Object-based Storage Systems, the performance, availability and scalability of the Metadata Server (MDS) cluster are critical. Traditional MDS cluster suffers from frequent metadata access and metadata movement within the cluster. In this paper, we present a new method called Hashing Partition (HAP) for MDS cluster design to avoid these overheads. We also demonstrate a design using HAP to achieve good performance of MDS cluster load balancing, failover and scalability. 1