Research Of Load Balancing Strategy In Cloud Computing

With the development of information technology, Cloud Computing currently has become a prevalent computing mode and one of academic research hotspots, which has great impact on people’s daily life and production. With the rapid development of Cloud Computing, its scale is continually expending, and the amount of user tasks need to be processed is greater. Thus, how to make full use of the system resource, improve the overall performance of the system, and ensure the Service-Level Agreement of the users, which is one of the important problems need to be solved in Cloud Computing. Effective load balancing strategy is an important measure to solve this problem. This paper studies load balancing algorithms and mechanisms, and surrounds the load balancing issue in Cloud Computing, do the following major works:1. Based on the importance of load evaluation for load balancing algorithms, proposes a method which uses Group-decision Analytical Hierarchy Process (GAHP) to evaluate the load state of the service node. In this method, adopting geometric mean to compute corresponding elements of judgment matrixes structured independently by multiple experts, and then gets the group decision judgment matrix. That can reduce the subjectivity and one-sidedness generated by man-made judgment matrix, enabling load evaluation of service node more objective and accurate.2. Proposes a method which using HHGA-RBF Neural Network to predict the load of the service node, and creates the load prediction model. In the aspect of training RBF Neural Network, adopts Hybrid Hierarchy Genetic Algorithm (HHGA) which includes Hierarchy Genetic Algorithm (HGA) and Recursive Least Square (RLS), uses HGA to determine the structure, node center and width of the hidden layer, and uses RLS to get the weights between the hidden layer and the output layer. Those enable RBF Neural Network converge faster, more efficient and load prediction more accurate, thus get effective load prediction. The experimental results show that, the load prediction values based on HHGA-RBF Neural Network are similar to load evaluation values, and the average relative error is less than0.01. The effect of this way is better than BP Neural Network.3. Based on node load prediction, Proposes a new dynamic load balancing algorithm-DPWRR, which combines with weighted round robin algorithm, and uses predictive value generated periodically to get the weight of the node. This algorithm can retain the advantages of static weighted round robin algorithm, and void its shortcomings. The experimental results show that DPWRR can achieve a better effect of load balancing.