Probabilistic model and optimal reorganization of B/sup +/-tree with physical clustering

We consider a variant of B/sup +/-tree called the B/sup +/-tree with physical clustering (BC), which is exemplified by VSAM key-sequenced datasets. In this file, logically contiguous data buckets are physically clustered in a storage unit called a region in order to enhance the efficiency of sequential processing. As a consequence, records in a bucket are redistributed over time according to two different split algorithms: bucket split and region split. A stochastic model of the disorganization of BC files is constructed using order statistics. Database administrators of online transaction systems often need to reorganize BC files before a region split occurs. An efficient algorithm for determining the optimal reorganization time of BC files is developed. Analytic results are validated by simulation. In passing, we discuss guidelines for load-time configuration of BC files.