Towards detecting software performance anti-patterns using classification techniques

Abstract

This paper presents a non-intrusive machine learning approach called Non-intrusive Performance Anti-pattern Detecter (NiPAD) for identifying and classifying software performance anti-patterns. NiPAD uses only system performance metrics-as opposed to analyzing application level performance metrics or source code and the design of a software application to identify and classify software performance anti-patterns within an application. The results of applying NiPAD to an example application show that NiPAD is able to predict the One Lane Bridge software performance anti-pattern within a software application with 0.94 accuracy.

Performance problem diagnostics by systematic experimentation

“…Peiris et al propose a non-intrusive performance anti-pattern detection (NiPAD) approach (Peiris et al, 2014). Instead of instrumenting the target application, the NiPAD approach solely requires system-level metrics that can be monitored without instrumentation of the application's code (e.g.…”

“…The discriminant function is derived by applying different machine learning approaches whereby labeled scenarios are used for learning the scenarios that contain performance anti-patterns. In (Peiris et al, 2014) the authors demonstrate a proof of concept for the non-intrusive detection approach by means of the One Lane Bridge (OLB) anti-pattern. Due to the high level observation of the target system, apart from detecting that an anti-pattern exists, the NiPAD approach inherently lacks the ability of diagnosing the root causes of performance problems.…”

“…Due to the high level observation of the target system, apart from detecting that an anti-pattern exists, the NiPAD approach inherently lacks the ability of diagnosing the root causes of performance problems. Moreover, from the work in (Peiris et al, 2014) it does not become clear whether the NiPAD approach is applicable to other anti-patterns than the OLB. Grechanik et al propose another approach that applies machine learning techniques for identification of performance bottlenecks (Grechanik et al, 2012).…”

Leveraging machine learning and information retrieval techniques in software evolution tasks

“…The second paper "Towards Detecting Software Performance AntiPatterns Using Classification Techniques" [13] was presented by James H. Hill from Indiana University-Purdue University Indianapolis. The paper described the NiPAD technique, which was devised to detect occurrences of a particular performance antipattern.…”

“…Techniques proposed in these fields could be used in building automated tools to support software evolution. Machine learning techniques include classification techniques that could be used to categorize bugs into families [3,19], predict duplicate bug reports [5,22], predict severity and priority of bug reports [11,20,21], predict effectiveness of fault localization techniques [6], predict the occurrences of defects or anti-patterns in software systems [8,10,13,17], and many others. Information retrieval techniques could be used to retrieve relevant documents from a large document corpus.…”