A Modeling Tool for Workload Analysis and Performance Tuning of Parallel Database Applications

Abstract

After about two decades of evolution parallel database architectures have finally reached a maturity level that makes them a serious competitor to traditional mainframe based architectures and distributed database systems. Dealing with these innovative systems requires to change considerably the application design process, mostly because of the sophisticated physical data organization (e.g. relation declustering), and, in general because of the more complex execution model. In this paper we present a methodology, and a connected modeling tool for the performance analysis of parallel database applications. The methodology is based on the results of previous work we have performed on this subject for traditional sequential DBMS, and is devised for a strict integration into the design procedure. The modeling approach allows both to provide the designer with an early feedback on the performance of the application, and a to produce a detailed account of the execution cost, that helps focusing the problem and gives a guide for the design improvement.