Measuring and Dealing with the Uncertainly of SOA Solutions

摘要

The paper investigates the uncertainty of Web Services performance and the instability of their communication medium (the Internet), and shows the influence of these two factors on the overall dependability of SOA. We present our practical experience in benchmarking and measuring the behaviour of a number of existing Web Services used in e-science and bio-informatics, provide the results of statistical data analysis and discuss the probability distribution of delays contributing to the Web Services response time. The ratio between delay standard deviation and its average value is introduced to measure the performance uncertainty of a Web Service. Finally, we present the results of error and fault injection into Web Services. We summarise our experiments with SOA-specific exception handling features provided by two web service development kits and analyse exception propagation and performance as the major factors affecting fault tolerance (in particular, error handling and fault diagnosis) in Web Services.