Adaptive Segmentation Schemes for Large Relational Database Machines

Abstract

The file segmentation is inevitable to cope with large files of information even in the design of database machines if we want to enlarge their capacity. The recent researches on database machines have much improved the processing speed by the introduction of special hardware algorithms for the search and the sort of relations. However, these are the algorithms for the internal processing of relations. In other words, these algorithms can not cope with such a relation whose size is much bigger than the buffer memory size. If a relation has too many tuples to read out all of them from the secondary memory into the primary memory or the buffer memory, the database machine has to divide this relation into a set of smaller segments so that each of them may fit into the primary memory size. If files are segmented arbitrarily, most queries require accesses to all the segments, which severely abates the system performance.