An Empirical Study Of BPEL Service Portfolio Data Flow Errors

With the booming of Web services and cloud computing,services composition has becoming one of the mainstream application development models.The Web Services Business Process Execution Language(WS-BPEL or BPEL for short)has becoming the de facto standard for describing the Web services composition.Since the BPEL services composition(process)is naturally block-structured,control flow related errors(eg,deadlocks)are not easy to occur in BPEL processes.In contrast,Web services interact with each other through messages.In order to ensure the privacy of data,BPEL services composition often needs to convert a large number of internal or external data,so it is very easy to introduce data flow errors,which seriously affect the quality of the BPEL services composition.Existing methods detect data flow errors which suffer from the trace explosion problem and bogus errors in BPEL processes.To this end,We propose two methods for detecting data flow errors in BPEL processes:a data flow analysis method based on the reaching definition and a symbolic encoding approach based on a satisfiability modulo theory(SMT)constraint solver.We apply the above two methods to 178 BPEL processes in industry,and make an empirical analysis of the data flow errors,and propose a data flow error classification and prediction based on the complexity metrics.This article mainly makes the following key contributions:1.We propose two methods to detect data flow errors in BPEL processes.The first one is a data flow analysis method based on the reaching definition.The second one is a symbolic encoding approach based on a SMT constraint solver,a symbolic encoding is firstly done for three common data flow errors,and then the SMT constraint solver is utilized to detect them.The first method can easily locate the location of the errors,but it suffers from false positives.The second method can ensure the feasibility of traces,but we need to encode and run the SMT solver.Therefore,the advantages and disadvantages of the two methods are complementary2.We make an empirical analysis of the common complexity metrics and the data flow errors of 178 real-world BPEL processes.The existing metrics are not enough to measure the data flow complexity and a set of new metrics is proposed to measure it.The statistical results show the seriousness of the data flow errors.There are a total of 94 BPEL processes containing data flow errors,accounting for 52.81%,of which the error of the output redundancy is the most common.We finally summarize the activity distributions and occurrence reasons of the data flow errors.3.The complexity metrics are used as the features to classify and predict the data flow errors.First,the method of selecting the candidate features from metrics is given,and then we do the classification and prediction for the data flow errors in BPEL processes based on various data mining classification algorithms.The accuracies of classification and prediction are all above 90%and the final selected features are simple and easy to calculate.