WEB SERVICE SELECTION BASED ON QoS PARAMETERS

Abstract

The potential of web service to automate machine to machine interaction has developed faith in using web service technology for building smart applications. It leads to increase in the interest of software developers to offer services in place of standalone software to the community of users. Multiple services offering identical functionality are available due to unabated uptake use of web service technology. In this situation, either the end user should manually select the web service or the system should have capability to identify the required web service satisfying the requirements of the user. The manual selection is not time efficient and does not results in the selection of optimal web service (low accuracy). In the other case, the selection process can be automated to identify the appropriate web service based on QoS or other parameters. With the enormous increase in number of web services offering identical functionality, the task of selection of web service itself become more difficult and challenging. The task of selection of appropriate web services for the case of composition becomes more challenging. The challenge is in terms of selection of desired web services by maintaining the system performance along with due consideration to the end-to-end QoS. The performance of web service selection mechanism has an effect on the performance of system built using web services. Poor performance of a web service selection mechanism may result into poor and inaccurate system performance. Also, mostThe majority of the existing approaches to web service selection consider all of the candidate web services for selection without any preprocessing or filtering. Thus, these existing approaches do unnecessary processing for those web services which are far below the expectations of the end user. The extra processing leads to poor selection results and degraded system performance. Further, web services run in a very dynamic environment and are prone to failure or run time change in service quality. On the occurrence of failure or unavailability of any component service, the reselection of services is required. In case of composite web service, the reselection process is more complex and time consuming. In this thesis, we have tried to handle these issues. We have studied the existing approaches to deal with above discussed issues and proposed improvements for selection of web services. A classification based approach for web service selection based on QoS parameters is presented. Initially, we evaluate the performance of eleven classification techniques for classifying web services. This analysis helped us to identify iv top three classification techniques for achieving efficient classification of web services. Majority vote based classification model is developed by combining the output of top three classification techniques and is used to classify web services. The classification step is used to prefilter the candidate web services. An improved PROMETHEE method, we call it as PROMETHEE Plus, is presented and is applied to most eligible web services for selection of most suitable services. Maximizing Deviation Method based hybrid weight evaluation mechanism is adopted for weight evaluation of various QoS parameters. We explore the use of clustering to determine similarity among candidate web services on the basis of QoS information. To identify the potential clustering technique, a systematic analysis is done to evaluate the performance of six clustering techniques based on efficiency and quality of clustering consideration. The best performing clustering technique is applied on candidate web services. Pruning along with clustering act as prefilter for candidate web services and this step identifies most promising set of web services with capability to meet the end user expectations. On the most prominent set of web services, proposed Skyline Plus method is applied for selection. Further, this thesis presents an efficient solution for replaceability of web services. Service selection is performed using modified genetic algorithm. To deal with the problem of service failure at run time, the replacement services are determined. In the presented approach, a web service selection mechanism by taking into cognizance the replaceability of service as the basis of selection is used. At the time of selection, the replaceability of a web service is determined using the proposed PROMETHEE Plus method. Further, the replaceability evaluation is enhanced by including IOPE based web service matchmaking. The matchmaking process is enhanced by using determinant method. We have gone through the existing works and found two appropriate and highly popular labeled datasets in this area. One of the dataset is QWS dataset which is very popular QoS dataset measured on real-world web services with 364 labeled web services and 2507 unlabelled web services. To the best of our knowledge, this is the only available dataset consisting of a labeled set of QoS information measured on real-world web services. This dataset is highly used for experimentation in the area of Web Services Selection. Another dataset is generated using synthetic data generator, which is highly popular and highly cited. We have performed experimentation on this dataset as well to validate our results. v Based on the findings of the works presented in the thesis, it is observed that the use of combination of classifiers improves the classification results as compared to using individual classifiers for classification of web services. The efficiency of the selection system is improved by the use of classification mechanism as a prefilter to the selection process. The results of experimentations and Tukey test showed that the inclusion of end user request during evaluation of services by PROMETHEE Plus method generates improved selection results with enhanced end user satisfaction. Further, it is revealed from the presented works that the quality of clustering and stability measurement of clustering techniques are important parameter to find the suitability of clustering techniques. The use of clustering technique to identify QoS based similarity among web services along with pruning act as an excellent prefilter for candidate web services. Use of pruning and clustering based Prefilter ensure that the final set of skyline services closely meet the end user expectations. The results from various experiments and Tukey test confirm that Skyline Plus gives improved selection results. It is further concluded that the selection of web services with consideration to replaceability improves the system availability and fault tolerance capability.