Automatically Finding Abstractions for Input Space Partitioning for Software Performance Testing

The goal of performance testing is to uncover problems where an application unexpectedly exhibits worsened characteristics for a speci c workload. It is di cult to construct e ective performance test cases that can nd performance problems in a short period of time, since it requires test engineers to test a large number of combinations of actions and data for large-scale applications. A fundamental question of performance testing is how to nd "key abstractions" that allow testers to select a manageable subset of the input data for test cases without compromising the e ectiveness of testing. We o er a novel solution for Abstraction Search for Input partitioning for Software performance Testing (ASSIST) for nding key abstractions for input space partitioning for performance testing automatically. ASSIST is an adaptive, feedback-directed learning testing system that starts with a small subset of test cases to enable testers to steer towards challenging tests automatically to nd more performance problems in applications in a shorter period of testing time. We have implemented ASSIST and have applied it to a dummy web application called JPetstore which has all the functionality found in any e-commerce application.