Achieving Predictability and Service Differentiation in Web Services

摘要

This paper proposes a model and an admission control algorithm for achieving predictability in web services by means of service differentiation. We use real-time scheduling principles typically used offline, adapt them to web services to work online. The proposed model and algorithm is empirically evaluated by implementing it Apache Axis2. The implementation is benchmarked against the unmodified version of Axis2 for various types of workloads and arrival rates, given different deadlines. We meet 100% of the deadlines keeping a healthy request acceptance rate of 42-100% depending on the task size variation. Our solution outperforms Axis2, specially at instances with high task size variance, by a factor of 10 - 1000.