Towards pro-active adaptation with confidence: augmenting service monitoring with online testing

  • Authors:

  • Andreas Metzger;Osama Sammodi;Klaus Pohl;Mark Rzepka

  • Affiliations:

  • University of Duisburg-Essen, Schützenbahn, Essen, Germany;University of Duisburg-Essen, Schützenbahn, Essen, Germany;University of Duisburg-Essen, Schützenbahn, Essen, Germany;University of Duisburg-Essen, Schützenbahn, Essen, Germany

  • Venue:
  • Proceedings of the 2010 ICSE Workshop on Software Engineering for Adaptive and Self-Managing Systems
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Service-based applications need to operate in a highly dynamic and distributed world. As those applications are composed of individual services, they have to react to failures of those services to ensure that the applications maintain their expected functionality and quality. Self-adaptation is one solution to this problem, as it allows applications to autonomously react to failures. Currently, monitoring is typically used to identify failures, thus triggering adaptation. However, monitoring only observes failures after they have occurred, which means that adaptation based on monitoring is reactive. This can lead to shortcomings like user dissatisfaction, increased execution times, and late response to critical events. Pro-active adaptation addresses those shortcomings, because in such a setting, the application detects the need for adaptation and thus can adapt before a failure will occur. However, it is important to avoid unnecessary pro-active adaptations, as they can lead to severe shortcomings, such as increased costs or follow-up failures. This means that when taking pro-active adaptation decisions it is key that there is confidence in the predicted future failures, i.e., pro-active adaptation should only be performed if there is certainty that the failure could in fact occur. To avoid unnecessary adaptations, we introduce an approach based on augmenting service monitoring with online testing to produce failure predictions with confidence. We demonstrate the applicability of our approach using a scenario from the eGovernment domain.