Linear hashing based dynamic file structures for non-uniform record distributions

Abstract

When a database system involves volumes of data too large to be stored in main memory, the database system suffers a well-known problem of I/O bottleneck. One of the promising solutions is to reduce search space and consequently the number of I/O requests. Hashing methods have been found to be one of the useful access methods for physical database systems implementation. In recent years, several methods have been proposed to extend the applicability of hashing methods to dynamic files. Particularly, linear hashing methods have better random access performance than any other methods. Also, it has an advantage over other dynamic methods in that it lacks a directory. But random accessing is possible but sequential accessing is not. By order preserving linear hashing, sequential accessing is possible but not always efficient. One difficulty is that if the natural order of each attribute field is to be preserved, the structure becomes dependent on the distribution of the keys in the key space. We propose a variant of linear hashing which is weakly order preserving and less depends on data distribution, where ordered records are redistributed in a multi buckets. Our method called redistribution order preserving linear hashing (ROPLH) uses a multi bucket segment for the purpose of redistribution. The objective is the development of file organizations that, with reasonable insertion cost, keep their optimal performance characteristics for random access even when the file size increases. Also, the file organization must provide for efficient range search. It will efficiently support a random search that is a part of an ordered searchings.