Rohr, Matthias, van Hoorn, André, Giesecke, Simon, Hasselbring, Wilhelm and Matevska, Jasminka (2008) Trace-Context Sensitive Performance Models from Monitoring Data of Software Systems . [Paper] In: Workshop on Tools Infrastructures and Methodologies for the Evaluation of Research Systems (TIMERS'08) at IEEE International Symposium on Performance Analysis of Systems and Software 2008. , April 20, 2008, Austin, TX, USA . Proceedings of the Workshop on Tools Infrastructures and Methodologies for the Evaluation of Research Systems (TIMERS'08) at IEEE International Symposium on Performance Analysis of Systems and Software 2008 . ; pp. 37-44 .

Preview

Text TIMERS2008.pdf Download (386kB) | Preview

Abstract

Operation response times in software systems are
typically modeled by probability distributions. However, particularly in Java EE applications, operation response time distributions are often of high variance or multi-modal. Such characteristics reduce confidence or applicability in various statistical evaluations. We observed that calling-context information, e.g., the complete call path within the system, is often connected to a significant part of this variance and other undesired distribution characteristics.

This paper introduces an approach to analyzing operation response times in the context of the complete call trace. This results in response time distributions that are specific to trace-contexts. We present empirical results of a medium-size online store demo application on the benefits of using trace-context specific response time distributions. The results are compared to the use of other or no calling-context information.

Empirical support is presented that trace-context analysis can create response time distributions with lower variance compared to using less or no calling-context information. Based on trace-context analysis, multi-modal distributions could be replaced by multiple unimodal distributions.

Actions (login required)

View Item