Study On Improved Cloud Service Selection Based On QoS Historical Data

Cloud computing came into being to satisfy the urgent need for computing power, efficiency of resource use and resource centralization. It is a computing model which offers the information resources, computing resources, storage resources and software to users through the Internet. With the substantial increase in the number of cloud services on the Internet, many cloud service providers offer a lot of services those have similar functions but different performance and price. In this context, how to help users find the most suitable service effectively is a very important task.There are three decision-making methods named Multi-criteria Decision Making(MCDM) method, Optimization-based method and Logic-based method to search the best service in several candidate services.Existing cloud service decision-making methods have some shortcoming, if they only consider the real-time QoS performance, the methods may lead to the selection of a service at local maxima; And if they only consider the average historical QoS performance, the methods do not capture the frequent variation in the QoS performance. Therefore Cloud Service Selection Based on QoS Historical Data has been proposed. This method can capture the frequent variation, it divided the historical time into equal and non-overlapping periods, then calculate the result of service dicision-making in each period(using MCDM method), and combine this result with the time weight model to calculate these candidate services' overall performance.For further analysis of the frequent variation in the QoS performance, the existing Cloud Service Selection Based on QoS Historical Data is improved. Firstly, in the original algorithm the criteria weights of service decision-making in each period are calculated by entropy method based on the average QoS historical data. In this paper, to further capture the frequent variation in QoS performance, the process to obtain the criteria weights is improved, we prefer to calculate the weights by entropy method base on the period's QoS data rather than base on the overall average QoS historical data. Secondly, use time series forecasting method to forecast the future QoS data, these data can be added into the original data set to generate a new data set, then service decision-making is done based on the new data set. These two improvements are designed to improve the accuracy of service selection. Finally, in order to evaluate two improvements, four experimental models are proposed to experiment progressively, then analyse the performance of the improved algorithm base on the comparison of experimental results.