A sequential pattern based QoS prediction method for web service

Web service composition (WSC) assumes that the service-level agreements (SLAs) the service provider commit remain steady throughout the composition's execution. However, WSC often operates in environment where the QoS of Web services changes relatively frequently. In this situation, the runtime QoS of real world WSCs will not always be equal to the QoS parameter settings specified by its SLA. Unfortunately, most existing researches still employ a static prediction strategy in predicting the QoS of WSC, where the QoS parameters published by SLA (or service providers) are simply trusted and used as model input for calculation of dynamic QoS metrics, and no runtime fluctuations of QoS parameters are considered. Instead of dynamic models, we propose a sequential pattern based method for dynamic prediction QoS of web services (SPPM). In our approach, the runtime QoS values of web service are interpreted as a time series model and QoS patterns for these data are defined. A transition matrix for QoS patterns is also calculated. The long-term future QoS value can then be predicted through a series of calculation on the transition matrix. Experiments show that the proposed approach achieves satisfactory accuracy.