Implementation of a Virtual Time Synchronizer for Distributed Databases on a Cluster of Workstations

Abstract

The availability of high speed networks and improved microprocessor performances has made it possible to build inexpensive cluster (or network) of workstations as an appealing vehicle for parallel and distributed computing. In this paper, we study the performance of a distributed database system synchronized, by virtual time (VT) mechanism by experimenting on a LAN connected collection of Sun SPARC workstations. To the best of our knowledge, ours is the first attempt to report the virtual time performance in distributed databases using a cluster of workstations. The experimental results demonstrate that the VT synchronization approach is a viable concurrency control method. Compared to a widely used multiversion (MV) concurrency control scheme based on timestamps order, the VT synchronization scheme is shown to yield about 25-30% reduction in the response time. The reason for choosing multiversion concurrency control as the basis for comparison is that both VT and MV protocols are based upon the concept of timestamps, and attain high performance by using multiple version of data objects. (The term version refers to the value of data item produced by a transaction that's either active or committed.)