Run-time performance testing in Java

This work focuses on relative comparisons of individual method's performance. It is based on Stochastic Performance Logic, which allows to express, for example, that one method runs at most two times longer than another method. These results are more portable than absolute values. It extends standard unit tests with performance assumptions, which are evaluated during actual run-time of a released application. Dynamically added and removed instrumentation is used for automatic modification of the production code. Instrumentation part uses DiSL framework to be able to seamlessly measure even Java system classes. Methods are measured sequentially, the number of concurrently measured methods is dynamically changed and the measurement code is removed as soon as required data are obtained. The results show that this approach may bring appreciably lower overhead peaks than measuring all methods at once. The prototype was compared with JMH tool and the results show that it is able to accurately measure methods running longer than 1 ms. Powered by TCPDF (www.tcpdf.org