Frotscher, Tom (2013) Architecture-Based Multivariate Anomaly Detection for Software Systems. (Master thesis), Kiel University, Kiel, Germany, 107 pp.

Preview

Text Master Thesis Final.pdf - Published Version Download (2MB) | Preview

Abstract

With growing software systems appropriate monitoring concepts become more and more important. For large-scale software systems this becomes a problem, simply because of the overwhelming size and complexity these systems can reach. Performance data is of special interest in this context, since it is related to factors such as the confidence and satisfaction customers have in the offered services.

In many cases, performance issues manifest themselves in so-called anomalies, that do not fit to the normal behavior of the software system. OPADx, an approach and a corresponding implementation, that is developed in the process of this thesis, is able to recognize such anomalies. In contrast to often used offline analysis approaches, the OPADx approach is able to process performance data online, which makes fast reactions to the anomalies and the correlated issues possible.

OPADx is based on an approach called OPAD, that was presented by Bielefeld in 2012. An analysis of OPAD reveals weaknesses of the approach, whenever anomalies are occurring over a long time period. Based on this analysis, extensions are developed and implemented to improve the approach. With the help of the implementation, the extended approach, then named OPADx, is evaluated on past and actual data of the social network Xing. Xing serves as case study environment for the actual thesis and also for the work of Bielefeld, which makes the results comparable. It turns out, that the extensions lead to improved anomaly detection results and an enhanced usability of the approach.

Actions (login required)

View Item