Impact-Driven Regression Test Selection for Mainframe Business Systems

Abhishek Dharmapurikar, Benjamin J. R. Wierwille, Jayashree Ramanathan, Rajiv Ramnath

Abstract: Software testing is particularly expensive in the case of legacy systems involving mainframes. At the same time these systems are critical to many large enterprises and they are perpetually in costly maintenance. For example, even small changes to the system usually lead to an end-to-end regression test. Also, due to the age of legacy systems there is a lack of essential knowledge (e.g. component inter-dependence) and this results in comprehensive system tests that have to be conducted in production environments. This is called the “retest-all” approach and is done to ensure confidence in the functioning of the system. But this approach is also impractical primarily due to: a) resource needs, and b) user stories generated within the agile processes that require changes to the system at an ever-faster pace. The research reported here is aimed at reducing the required regression testing and the costs associated with the system and its assets. The improvements are achieved by identifying only those tests needed by assets changes and others that are ‘impacted’. The impact analysis leverages modern static code analysis tools such as Rational Asset Analyzer and dedicated test environments for mainframes. We show that by using impact analysis on a real-world mainframe application the test savings can be about 34%.