In parallel object-oriented database systems based on shared-nothing parallel processing architectures, object placement among the nodes has a significant impact on response time and throughput. This is because inter-nodes object references and skewed workload distribution among the nodes are serious problems for the effective utilization of parallelism.
In this thesis, we propose a two-phase object declustering algorithm for parallel object-oriented database systems. Here we mainly focus on minimizing the number of inter-node object references and achieving uniform workload distribution across the nodes. In the first phase, we construct object clusters for each class-composition hierarchy. This is to minimize inter-node object references. In the second phase, we allocate object clusters to the nodes in such a way that appropriate load balancing among the nodes can be achieved. We show through performance experiments that the query response time is better than the greedy object declustering algorithm proposed earlier.