Distance Based Root Cause Analysis and Change Impact Analysis of Performance Regressions

Abstract EN

Performance regression testing is applied to uncover both performance and functional problems of software releases.

A performance problem revealed by performance testing can be high response time, low throughput, or even being out of service.

Mature performance testing process helps systematically detect software performance problems.

However, it is difficult to identify the root cause and evaluate the potential change impact.

In this paper, we present an approach leveraging server side logs for identifying root causes of performance problems.

Firstly, server side logs are used to recover call tree of each business transaction.

We define a novel distance based metric computed from call trees for root cause analysis and apply inverted index from methods to business transactions for change impact analysis.

Empirical studies show that our approach can effectively and efficiently help developers diagnose root cause of performance problems.