Evaluating the DSMIO cache-coherence algorithm in cluster-based parallel ODBMS

In this paper, we assess the performance of DSMIO cache-coherence algorithm implemented in a parallel object-based database management system (ODBMS). The distinguishing feature of DSMIO is its use of the lazy release memory consistency model and multiple-writer protocol to reduce both the number and size of coherence messages required to keep coherent a distributed ODBMS across a cluster of PC servers. Using a large distributed database and several application workloads we evaluate DSMIO performance and also compare it against that of the well-known Call-Back Locking (CBL) algorithm. Our results show that both algorithms perform very well for read operations whereas DSMIO outperforms significantly CBL for write operations with DSMIO speed-ups attaining as much as 5.4 while CBL speed-ups reach at most 1.4 for an 8-node cluster. Overall, these results suggest that designers of cluster-based ODBMS should consider DSMIO as an efficient option for developing future projects in the field.