Research On Optimizing Services Composition Based On T-QoS-awareness

With the increasing of the cloud users’ requirements, a single cloud service has been unable to meet the complex requirements of cloud users. Service composition combines the existing distributed cloud services into composite services which can meet cloud users’ requirements. However, it is an urgent problem how to construct a composite service, which both meet the complex requirements of cloud users and has the global optimal quality. However, the existing research results have the following shortcomings.1) For the selection of each atomic service in the service composition, the existing approaches only consider the quality of service, but ignore the credibility of service, which lead the failure of the execution of composite services for the existence of malicious service entities.2) The initial distribution of pheromone is set to be constant in the traditional ant colony algorithm, which make it easy for the ants to choose high-quality services in the first iteration and make ants choose the low total quality composition as the optimal solution. Finally the initial search will be slow.3)The traditional ant colony algorithm uses the quality of the selected service as the size of pheromone when updating the pheromone on the optimal path. But the quality of some atomic service of the optimal service composition could be small, which lead the ants to choose the high-quality service on the sub-optimal path and miss the optimal service composition path.According to the problems above, a Research on Optimizing Services Composition based on T-QoS-awareness is proposed. Specific work and innovation made as follows:A dynamic trust evaluation model based on evaluation credibility is proposed firstly. This model divides the ability of cloud service provider and the one of the user’s requirements into many ranks, which can effectively solve the potential damage caused by the dynamic change in the ability of cloud service providers. A dynamic mechanism of trust changing about time-window is established. During the calculation of credibility, the user’s evaluation credibility is used as the trust weight. The calculating accuracy of the recommended behavior credibility is improved by introducing the evaluation credibility and evaluation similarity. Simulation results show that the model result is closer to the cloud service provider’s actual trust value, and can resist the attack of malicious cloud users effectively.Secondly, a risk prediction model based on improved AdaBoost method is proposed. In this thesis, we regard the risk prediction as two-class classification problem, and predict the risk of new cloud users through the attribute of historical cloud users. In order to classify precisely, AdaBoost method is adopted in this thesis. This method through the error rate of the last training to adjust the sample distribution of the next training, make the next training has stronger ability of identification for the error samples, and reset a weight for each weak classifier. Last, through the means of weighted voting to generate the strong classifier, thus improve the overall results of classification. Considering the wrongly-predicted cost, AdaBoost method is improved in this thesis, introduce the cost-sensitive, and consider the cost-sensitive into the structure of the classifier, thus generate the minimal cost classifier. Experiments showed that, the cost-sensitive AdaBoost method has better effect than the traditional one, thus the method can predicts the risk of the new cloud user effectively, and protect the security of the cloud services effectively.Finally, a T-QoS-awareness based parallel ant colony algorithm is proposed. Our approach takes the service credibility as the weight of the quality of service, then calculates the trust service quality T-QoS for each service, making the service composition in a credible environment; Through the establishment on a per-service T-QoS initialization pheromone matrix, we can reduce the colony’s initial search time; By modifying the pheromone updating rules and introducing two ant colonies to search from different angles in parallel, we can avoid to fall into the local optimal solution, and quickly find the optimal combination of global solutions. Experiments showed that, our approach can find a combination of high-quality services while improving the operation success rate of the services at the same time. Also, the convergence rate and the accuracy of optimal combination are improved.