The Research And Optimization Of Cloud Resource Allocation Model Based On SLA

Cloud computing is a new way of Internet-based computing, it can provide resources in accordance with customer’s demand. In recent years, cloud computing has become the mainstream of the development trend of the information industry, and also become a hot academic research. But with the expansion of the scale of the user, the user’s request showing increasing diversity, dynamic, complex features, and as well as the energy consumption of data center has become increasingly prominent, these all bring challenges to the scheduling assignment of cloud resources. At the same time, users’requirements of quality of service such as response time are getting higher, while SLA can provide a guarantee of quality of service. Therefore, the research of resources allocation based on SLA is necessary.Based on the analysis of current research, this thesis presents an improved cloud computing resource allocation model, and adoptes a swarm intelligence algorithm for its solution. The main contents are as follows:(1) This model fully considers the needs of users on the quality of services, meanwhile establishes the objective function on the goal that cloud service providers can obtaine maximize profits. The function modules in the cloud resource management platform are introduced first, and the flow of information between them are analyzed. Second this model uses queuing theory to model the process of requests are served and evaluate the response time, and uses a layered approach to handle users’requests, considers the load status of cloud computing system, according to the predicted number of requests to decide the allocation of virtual resources.(2) Multiple variables with binding relationship exist in the above resource allocation model, and it belongs to NP-hard problem. Traditional solution is divide this problem into several sub-problems by greedy algorithm for solving the problem, but the results of sub-dividing obtained after summing need to prove is the optimal solution of the original problem, and the prove is a complex process. Therefore, this theis uses swarm intelligence algorithm to solve the model. The particle swarm optimaziton algorithm as one of intelligence algorithm has easy implementation, high accuracy, fast convergence, and can complete the whole coding of all variables. Therefore, this thesis uses this algorithm to solve the model. And in the process of solving the problem the method of the continuous processing of discrete particles are introduced.(3) Taking into account that when the population size is too large, or when the number of requests relatively many, it will take a long time to slove the model on a single server, so a method of parallel solve based on distributed programing model MapReduce is presented.Finally, experiments for solving the model are made on a stand-alone environment and Hadoop-based distributed platform, and the experimental results are compared and analyzed. Experiments show that, PSO algorithm can be more effective to find the optimal solution dynamics, and in the larger population, the parallel PSO algorithm can significantly save computing time and improve efficiency.