Index clustering for high-performance sequential index access

Abstract

This paper presents an index clustering technique called the segment-page clustering (SP-clustering). Most relevant index pages are widely scattered on a disk due to dynamic page allocation, and thus many random disk accesses are required during the query processing. The SP-clustering avoids the scattering by storing the relevant nodes contiguously in a segment that contains a sequence of contiguous disk pages and improves the query performance by offering sequential disk access within a segment. A new cost model is also introduced to estimate the performance of the SP-clustering. It takes account of the physical adjacency of pages read as well as the number of pages accessed. Experimental results demonstrate that the SP-clustering improves the query performance up to several times compared with the traditional ones with respect to the total elapsed time.