Research On Virtual Resource Scheduling Method Under Cloud Environment

In recent years, cloud computing has become a research hotspot in the field of business and academic, and develops rapidly. As a new type of commercial calculation model, the core idea of cloud computing is to distribute a large number of computing tasks on the resource pool which is made up of cheap computer cluster so that applications can request an on-demand access to computing power, storage space, and information services. However, in the face of a huge number, strong distribution and dynamic change of resource, which method is adopted to improve the resources of the organization and scheduling is one of the key technologies for cloud computing. Most of the existing resource scheduling methods consider the task as a whole, no for task on resources the property specific requires for effective analysis and not fully consider the load indicators parameter during the resource scheduling process. Furthermore, these resource scheduling in process of optimization genetic algorithm are no too much to concern the improvement of selection operator and crossover operator, making algorithm vulnerable to local optimal problem and leading to potential excellent gene of lost.Aiming at optimal span and load balancing problems in a cloud environment, an intelligent optimization strategy about virtual resource scheduling problem is proposed in this thesis. Take the characteristics of cloud virtual resources into consideration, this strategy analysis the constraint indicators of cloud computing resource scheduling optimization model. Both optimal span and load balancing are measured as constraint indicators in resource scheduling process. The definition of optimal span indicator is the sum of latency, transmission time and execute-time of the task. And the load balancing indicator is defined as the comprehensive utilization of CPU, bandwidth and memory. In the evolutionary process, using a fitness function to measure the degree which individuals in groups may reach or close to the optimal solution in optimization computation. Larger fitness value indicates that there are more opportunities to breed the next generation. But from a user perspective, the completion time of task should be as short as possible in execution process. When it comes to the system performance, the scheduling algorithm should try to ensure the load balancing of nodes. Thus, taking the countdown of optimal span and loading balance as a dual fitness function in optimization of genetic algorithm. In addition, the selection mode uses optimum reserve replacement strategy. The strategy sorts the fitness function values of each individual in the selection process to pick out the best and worst solution in current generation and compared with the previous generation. If previous best value is larger, replace the current best value. Otherwise, replace the current worst value with the previous best value, the current best value does not join in selection mode. The crossover operator adopts the way of single point crossover and order cross combination, sets the crossover intersection between tasks to prevent the same task is assigned to the same resources and makes the cloud cluster resource scheduling process be able to recommend a good resource for processing to ensure that resources are load-balanced and increase scheduling efficiency.Finally, cloudsim simulation experiment is done in order to verify the validity of the algorithm. In the same environment conditions, comparing the experimental results with virtual resource scheduling method proposed in this thesis with the cloud resource scheduling of traditional genetic algorithm (GA) and the improved genetic algorithm (IGA), the results analysis show that the strategy can improve the efficiency of resource scheduling under the massive task, optimize resources load balance, so as to verify the effectiveness of the algorithm.