Block access estimation for clustered data using a finite LRU buffer

Data access cost evaluation is fundamental in the design and management of database systems. When some data items have duplicates, a clustering effect that can heavily influence access costs is observed. The availability of a finite amount of buffer memory in real systems has an even more dramatic impact. A comprehensive cost model for clustered data retrieval by an index using a finite buffer is presented. The approach combines and extends previous models based either on finite buffer or on uniform data clustering assumptions. The computational costs of the formulas proposed are independent of the data size or of the query cardinality and need only a single statistics per search key, the clustering factor, to be maintained by the system. The predictive power and the accuracy of the model are shown in comparison with actual costs resulting from simulations.<
>