Research Of Hierarchical Resource Allocation Model In Cloud Environment

With the information industry development, the era of big data is coming. The data challenges the network structure. Cloud computing is relying on virtualization technology, grid computing, distributed computing and collaborative computing. Cloud environment is heterogeneous and dynamic, so it is important to how to allocate resource.For multiple cloud nodes collaboration in cloud environment and resources dynamics, dynamically hierarchical network computing model is created, and a dynamically hierarchical resource-allocation algorithm (DHRA) is proposed. Fuzzy pattern recognition theory was introduced into dynamically hierarchical network computing model, which divided tasks and resources into different levels based on its information. Thereby the dynamically hierarchical network computing model is formed. Hence, each task only needs to search the right node in the corresponding level, the scale of the problem was reduced. On this basis, the introduction of multi-Agent in the resource-allocation algorithm increased the reliability and autonomy of the system. Integrated task completion time, node's load, the communication traffic and so on, DHRA gets better performance and efficiency in various aspects. As for the parallel computing of large application sub-tasks which are inter-related each other, since all the calculation of tasks and required resources are known, the genetic algorithm which is one kind of random-search algorithm is chosen. In order to optimize various performances, multi-object genetic algorithm (MOGA) is proposed. MOGA takes two fitness functions that refer completion time and task-node correlation to control the direction of evolution of the population. Communication traffic and completion time were reduced.In this thesis, quantitative analysis for the traffic of DHRA has been given, the result proved that DHRA can effectively reduce traffic. The simulations of DHRA and MOGA at different number of tasks and nodes have been done. Compared with MinMin algorithm, DHRA reduced communication traffic, and completion time was also reduced. Similarly, compared to GA, MOGA obtained less completion time and traffic in the same conditions. The proposed algorithms effectively improved the stability and efficiency of the system.