A QoS-Enabled WorkManager Model for Web Application Servers

Abstract

The overload of Web application servers (WAS) is a typical and critical problem encountered in nowaday Web commerce. This problem imposes greater demands on WAS to provide QoS support under overload conditions. However, even with a practical solution, the implementation of a flexible QoS model to enable finegrained QoS control in WAS is still a challenging task. In this paper we propose a QoS-enabled workmanager model (WMQ), which hides the complexity of QoS mechanisms implementations and provides a flexible QoS architecture. Hybrid QoS schemes are proposed base on this model to support self-optimization and fine-grained QoS control. We implement this model and integrate it in a Web application server. Our integrating experience shows that the WMQ model significantly reduces the implementation effort in developing a QoS-enabled WAS. We evaluate this model by a TPC-W workload generator in a typical e-commerce application. Evaluation results show that, our overload control allows consistent performance and improves the system throughput up to 22% during extreme overload. In addition, compared with other widely used scheduling policies, our hybrid request scheduling policy shows better performance on improving the system throughput and minimizing the number of aborted requests.