An efficient method to predict the execution of web services in composition by singular values decomposition (SVD)

In service computing, the principal factor is the existence of a reliable composite web service (WS) qua it imposes the minimum penalty when a failure occurs. The variant strategies have already been presented to encounter failures in the composition which have been categorized in two groups (forward and backward recoveries). Since the distinct recoveries have an individual influence on QOS factors, the recovery strategy should be dynamically selected in runtime, considering the user preferences and the attitude of the user towards the most important factors. Dependencies between services can cause tremendous failures in counting the execution of composition. In this paper, an algorithm has been presented which can deduce the data dependencies between the web services by using singular values decomposition (SVD). To evaluate the presented algorithm as a benchmark, four workflows of real web services with the degree of different parallels in the execution of services have been implemented in BPEL designer environment in eclipse. The experimental results show that the presented SVD algorithm can deduce a data dependency and the executive status of the rest of services when a failure occurs. The more the training set is large, the more the prediction of the executive status of web services will be accurate.

Keywords:

composite web service (CWS) , fault tolerance , singular values decomposition (SVD) , forward recovery , backward recovery