Abstract
Transactional database management systems (DBMS) have been successful at supporting traditional transaction processing workloads. However, web-based applications that tend to generate huge numbers of concurrent business operations are pushing DBMS performance over their limits, thus threatening overall system availability. Then, a crucial question is how to test DBMS performance under heavy workload conditions. Answering this question requires a testing methodology to set up non-biased conditions for pushing a particular DBMS over its normal performance limits (i.e., to stress it). In this article, we present a stress testing methodology for DBMS to search for defects in supporting very heavy workloads. Our methodology leverages distributed testing techniques and takes into account the various biases that may affect the test results. It progressively increases the workload along with several tuning steps up to a stress condition. We validate our methodology with empirical studies on two popular DBMS (one proprietary, one open-source) and detail the defects that have been found.