On the performance of B-trees using dynamic address computation

Abstract

<p>The B-tree is a one of the more popular methods in use today for indexes and inverted files in database management systems. The traditional implementation of a Bâ tree uses many pointers (more than one per key), which can directly affect the performance of the B-tree. A general method of file organization and access (called Dymanic Address Computation) has been described by Cook that can be used to implement B-trees using no pointers. A minimal amount of storage (in addition to the keys) is required. An implementation of Dynamic Address Computation and a B-tree management package is described. Analytical performance measures are derived in an attempt to understand the performance characteristics of the B-tree. It is shown that the additional costs associated with Dynamic Address Computation result in an implementation that is competitive with traditional implementations only for small applications. For very large B-trees, additional work is required to make the performance acceptable. Some examples of possible modifications are discussed.</p>