SFERA: A Simulation Framework for the Performance Evaluation of Restart Algorithms in Service-Oriented Systems

In service-oriented systems, fault detection and localisation are not straightforward, and client-side fault-tolerance techniques are required to reduce the impact of faults on the quality of service experienced by the user. Restart is a well-known client-side technique for improving performance and service availability. With restart, tasks whose completion-time exceeds a timeout are re-issued by the client, with the goal of obtaining a shorter completion-time on the next attempt. Evaluation of restart should be performed by a combination of analysis, simulation, and measurement. In this paper we present the SFERA framework for simulation of restart in complex SOA systems. We illustrate SFERA features with an evaluation of the optimal restart timeout in a complex SOA system. We simulate a SOA system using different scenarios and model component response-times by phase-type distributions fitted to measurements from a SOA testbed. We observe and compare completion times for different scenarios.