Enhancing SKA software testing through data mining strategies

The Square Kilometre Array (SKA) is a groundbreaking radio telescope project, aiming at constructing the two biggest radio telescopes in Australia and South Africa. They will have a larger collecting area and sky resolution than existing radiotelescopes, and they will handle an unprecedented amount of data flowing between computing facilities. The functionality of these telescopes heavily depends on the quality of the operating software. The project’s magnitude and complexity require effective testing processes capable of preemptively identifying and addressing potential bugs and errors. In this context, a simple regression testing strategy is not enough. In the first years of SKA construction, we noticed that tests, which typically pass, may occasionally experience failures. Collecting and analyzing test results over extended time periods could help in understanding the origin of such failures and to find solutions that address them. It would be a significant step forward to improve the reliability of SKA software. Data mining is a process of discovering patterns, trends, correlations, or useful information from large sets of data. It can be applied to a large set of test results concerning the operations of a specific SKA software component, i.e. the Local Monitoring and Control of Central Signal Processor (CSP.LMC). The CSP.LMC is tested with a multilevel strategy, spawning from unit to system tests, that can be performed on different environments. In this paper we analyze the strengths of this approach, describe some of the pitfalls in implementing it, and discuss the possibility to apply it to different SKA Software components.