Content area
Abstract
Database systems have become extremely popular in nearly every area of scientific, business, and personal computing. Users expect immediate access to data while increasing the load on the system. Many database software systems in place today are reaching performance limits due to I/O constraints. Parallel I/O is an alternative to serial I/O systems that can offer a partial solution this problem of scalability. In this paper we evaluate the use of parallel I/O in a heterogeneous computing environment for performance impact on database applications. We created a database system that uses parallel database servers and parallel I/O. PVM and PIOUS provide the parallel services. We then ran a suite of benchmark tests including various parallel configurations and number of clients. When we tested the database using both parallel I/O and parallel database servers, total benchmark run time of eight clients was only slightly longer than that of a single client. When we removed either parallel I/O or parallel database servers from the test, the run times increased incrementally with additional clients. The resulting analysis shows that, in certain configurations, the combination of parallel I/O and parallel database servers makes the database system more scalable than its serial I/O counterparts.
