Runtime Behavior Monitoring and Self-Adaptation in Service-Oriented Systems

Abstract

Mixed service-oriented systems composed of human actors and software services build up complex interaction networks. Without any coordination, such systems may exhibit undesirable properties due to unexpected behavior. Also, communications and interactions in such networks are not preplanned by top-down composition models. Consequently, the management of service-oriented applications is difficult due to changing interaction and behavior patterns that possibly contradict and result in faults from varying conditions and misbehavior in the network. In this paper we present a self-adaptation approach that regulates local interactions to maintain desired system functionality. To prevent degraded or stalled systems, adaptations operate by link modification or substitution of actors based on similarity and trust metrics. Unlike a security perspective on trust, we focus on the notion of socially inspired trust. We design an architecture based on two separate independent frameworks. One providing a real Web service test bed extensible for dynamic adaptation actions. The other is our self-adaptation framework including all modules required by systems with self-* properties. In our experiments we study a trust and similarity based adaptation approach by simulating dynamic interactions in the real Web services test bed.