A Study On Composite Service Optimizing Selection Approach Based On Execution Information

With the widespread using of service computing techniques, there will be more and more Web services with one same function, and composite services which can provide value-added functions are also getting more attention. Therefore, how to select a group of Web services among those services with same functions, making the composite service having best quality and highest user’s satisfaction, has become an exigent issue to be sovled.Currently, there have been many research works about composite service selection, and most of them are QoS based selection approaches. In theory, these approaches can select a composite service with best quality on the premise of satisfying user’s constraints. However, they face many practical problems in the real selection environment. At first, the accurate QoS data of candidate services can’t be provided for service selection using current service computing techniques, making the selection depended information be unaccurate. Secondly, there are complex correlations among atomic services in a composite service, and these correlations influence the performance of atomic services and further influence the quality of the composite service. In addition, each Web service has its own features, and has different performance in different environment conditions. So it is difficult to accurately measure the service with only one QoS value during the selection process. Current service selection approaches ignore the above issues, making the selected optimal composite service in theory be not optimal in the real environment, or even can’t be executed. Then, the actual availability of these approaches is decreased. To solve the above issues, an execution information based composite service optimizing selection approach is proposed in this paper. This approach uses some technologies such as data mining, and based on specific information collected during the execution of composite services, mines different kinds of knowledge patterns which can be used to do composite service optimizing selection. Then the discovered knowledge is used to optimize the selection results for the above problems. Surrounding the proposed approach, in-depth studies are done in the following aspects.(1) To acquire the related composite service execution information that the proposed approach uses to do mining, the techniques about Web service logging are studied, and an API Hook log recorder based common composite service logging framework is proposed. This framework can be easily deployed in the service composition systems, and collect composite service execution data flexibly.(2) For the problem that the selection depended candidate service QoS data is unaccurate, the execution log based Web service QoS acquisition mechanism is researched, and a composite service optimizing selection approach based on Web service TQoS Acquisition is proposed. There are2problems for Web service QoS acquisition based on practical execution. Firstly the collected data amount is small. Secondly, the collected data distributes unevenly. To solve these2issues, a density clustering based valid time span determining approach and a Lagrange interpolation based QoS point data filling approach are proposed respectively. Experimental results show that this optimizing selection approach can acquire the QoS data of candidate services accurately, then to do composite service optimizing selection effectively.(3) For the problem that there are complex correlations among atomic services which influence the composite service quality, the correlation considered composite service selection is studied, and a SESE decomposition based service correlation sensitive optimizing selection approach is proposed. This approach first discovers efficient SESE patterns frequently used in the past composite service execution instances, and then divides the composite service process into some SESE regions. Regarding the divided SESE regions as selection unit, the proposed approach can do wholly selection for all the services in one region, and it can improve the service selection performance..(4) For the problem that each Web service has its own features and it’s difficult to measure it using only one QoS Value, the candidate service feature aware composite service selection is studied, and a Web service feature analysis based service optimizing selection approach is proposed. In this approach, the concept of "feature rules" is defined to represent Web service features formally, and the feature rule mining algorithm is given. At last, the method about how to use Web service feature rules to do service optimizing selection is presented.(5) In real selection environment, the above practical problems usually exist at the same time. Considering how to integratedly use all optimizing selection approaches, an integrate application process of the3composite service optimizing selection approaches is given. This approach can combine the above optimizing selection approaches organically, and improve the performance of each service selection approaches in real service execution environment.(6) To verify the composite service optimizing selection approach proposed in this paper, a staged optimization experimental platform of composite service self-adaption is designed and implemented, and some experiments are done on it. The results show that the proposed3optimizing selection approaches can improve the performance of selected composite services step by step, and increase the practical usability of current selection methods which use them. So, the optimizing selection approach proposed in this paper is an effective solution to the practical problems that composite service selection faces.