An efficient method for distributing search structures

Abstract

The paper presents a method for distributing search structures which have an index in shared memory multiprocessor multi-disk systems. It is shown that in comparison with currently used distribution methods, the new method generates distributed search structures having a much lower main memory space requirement, similar parallel execution capabilities, a similar execution cost per operation, and a fairly close disk space utilization. Analyzing the disk space utilization of the resulting distributed structures is non-trivial. The paper briefly explains how it has performed such an analysis (which is partially based on a technique called fringe analysis). The result is that the benefits of the method are obtained at a very small cost in terms of reduced disk space utilization.<>