A pattern-based clustering strategy for object-oriented databases

Due to the navigational characteristic, queries in an object-oriented database system (OODBS) often access a series of related objects. A clustering strategy can be used to place related objects on the same disk page such that the number of I/O operations required for query processing can be reduced. Therefore, the clustering strategy is important for improving the performance of OODBSs. In this research, we propose a new clustering strategy which uses both the object affinity and access pattern information to cluster objects. We first use a set of queries to generate the training traces. Each query produces a graph that represents the access pattern of the query. Then, we use these patterns as clustering units to group objects. If the size of objects in one graph is larger than the size of one disk page, we partition the objects in the graph. Simulation results of our new clustering strategy are also included