Study on Dynamic Load Balancing Based on Load Predicition and Overload Migration
|School||Qingdao Technological University|
|Course||Applied Computer Technology|
|Keywords||Load Prediction Load Migration Genetic Algorithm Simulated Annealing Optimize Portfolio|
In today’s era of information explosion, new applications continue to emerge ,the traffic of data flow and calculation of the enterprise network、campus network or on the Internet intensity increased. The traditional single-server mode has apparently been unable to meet the high scalability, high availability network services to deal with the growing load capacity. Therefore a server clusters conformed with high-performance servers become an effective structure to achieve high scalability and high availability of network services.Cluster technology is to connect many independent servers and provide services with the external form of the whole cluster. Service requests must be assessed on every server to achieve the parallel running、shorten the access time and optimize the overall performance. Load balancing mechanism is the core issue of cluster technology.Effective load balancing mechanism can extend the server bandwidth and increase system throughput, but the influence of the node performance parameters in cluster server system and the dynamic and instability of the load make it very difficult to achieve the whole system load balancing use a simple single-task scheduling program. So the study on dynamic feedback load balancing algorithm based on load migration is proposed.The study in this article mainly includes two parts: the load prediction and load migration. With the feedback dynamic collection of the performance parameters data of the node of the cluster server, we use the ARMA(p,q) model to predict the node load status. Because there is a certain degree of the prediction model adaptiong and the load balancing after the predition, on the basis of node load prediction we introduced an improved simulated annealing genetic algorithm optimization to select the corresponding node to load migration as a supplement to load balance. With the more accurate prediction of the server node workloads, try to allocate the mission reasonablly to the service node, at the same time perform load migration strategy to the overload node, so each node will take a more balanced load, thereby enhancing the flexibility of the network server and its data processing capabilities, achieve the overall system load balancing, improve system quality of service.