Study On QoS-aware Service Selection

Web service has been a hot area of research in academia and industry, because ofits good interoperability, loose coupling, scalability, and many other advantages.Although the technology of web service becomes more mature, a single atomic webservice still cannot meet the users’ complex requirements. As a result, web servicecomposition turns into a common choice for many users. Making use of web servicecomposition, users can complete more complex requirements by dynamicallyintegrating existing services. Along with the competition increasing in the networkingindustry, web service composition should not only meet the users’ functionalrequirements, but also provide users the best user experiences. So it becomes a majorproblem that how to select the appropriate atomic services to ensure that the servicecomposition has the best quality of service (QoS), when the system is comprised of alarge number of services with the same functionality but different non-functionalproperties.When web services are invoked online, some QoS attributes will demonstrateunstable, such as the response time of the web services. In QoS-aware service selection,existing researches mainly take use of the static QoS which is provided by serviceproviders or the historical average value from service operations. However, neither thestatic value nor the historical average value can accurately reflect the real value ofresponse time during each service operation time interval, which will surely influencethe quality of service composition after service selection. To solve this problem, we usethe average value of web service response time in different sub-periods, and propose aweb service response time prediction method which is based on the Kalman predictor toforecast response time in the next period. The experiment shows that this method caneffectively forecast the response time when a web service is running.To select services for a service composition with multiple execution paths, existingmethods often select services for each path separately and then integrate all the resultsfor each path to get the final choice. These methods are not efficient. In order to solvethis problem, we provide a QoS calculation method based on Markov chain for webservice composition with multiple execution paths. Using this method, we can get thefinal result by only one selection process.Traditional service selection algorithms are time-consuming when there are huge numbers of atomic or abstract services in a composition. We can use genetic algorithm(GA) to solve this problem for better performance. The traditional genetic algorithm isnot good enough to solve such problem when the transmission cost among networks istaken into account. In this paper, we propose an improved genetic algorithm to getbetter quality of service selection by improving the initial population generationprocess.