Performance upgrade for large object bases using disk arrays

Abstract

The authors propose the use of multiple disks to increase the bandwidth of disk I/O so as to alleviate the I/O bottleneck problem existed in OODB applications. However, a major difficulty in using multiple disks for the support of OODBs is that an object may reference or be referenced by other objects through associations. This requires that the search of objects be sequential through the association links, which is sometimes referred to as the navigational search of objects in an object base. This restriction limits the concurrency in object retrieval and invalidates the use of multiple disks. Techniques such as clustering/indexing of objects do not help to resolve this problem. By using a two-phase query processing strategy previously proposed by them, objects of interest can be accessed in parallel from multiple disk. The aim of the paper is to design a probability-based evaluation model to estimate the performance of using multiple disks and to compare it with a single disk system. The result shows that the scheme is viable and a significant improvement is achieved while multiple disks are used.