Towards dependable Web services

Web services are the key technology for implementing distributed enterprise level applications such as B2B and grid computing. An important goal is to provide dependable quality guarantees for client-server interactions. Therefore, service level management (SLM) is gaining more and more significance for clients and providers of Web services. The first step to control service level agreements is a proper instrumentation of the application code in order to monitor the service performance. However, manual instrumentation of Web services is very costly and error-prone and thus not very efficient. Our goal was to develop a systematic and automated, tool-supported approach for Web services instrumentation. We present a dual approach for efficiently instrumenting Web services. It consists of instrumenting the frontend Web services platform as well as the backend services. Although the instrumentation of the Web services platform necessarily is platform-specific, we have found a general, reusable approach. On the backend-side aspect-oriented programming techniques are successfully applied to instrument backend services. We present experimental studies of performance instrumentation using the application response measurement (ARM) API and evaluate the efficiency of the monitoring enhancements. Our results point the way to systematically gain better insights into the behaviour of Web services and thus how to build more dependable Web-based applications.