An Improved Strategy For Load Balancing In Cluster Nodes

The rapid growth of the World Wide Web, the fast popularization of Broadband, the increasing quantity of netizen, and various web-based applications are creating demand for high performance web servers to offer better throughput and shorter user-perceived latency. This demand leads to widely used cluster-based web servers in the Internet infrastructure. Centralized cluster is more effective at load balancing than distributed cluster, but centralized cluster contains a serious weakness, system bottlenecking. This is because a front-end dispatcher existing in the cluster system is the only access point, which not only bears monitoring the load states from the back-end servers, but also must complete distributing the load among those servers. Therefore, when network traffic is large, it will easily become the bottleneck in the system, and even cause the collapse of the system in severe cases.In order to solve this problem, these tasks in the following are completed in this paper.First, a basic scheduling strategy of load-balancing based on principle of a discrete uniform distribution is proposed to solve the deficiencies of Weighted Round Robin load balancing algorithm in this paper. In this scheduling, each server sends the ask stream at different rates to the balancer and these streams can carry the load conditions, so the balance does not monitor the workload while the back-end servers do, and as a result is that the strategy achieves reducing the work pressure on balancer.Second, dynamic scheduling mechanism is introduced and used in the proposed strategy on load balancing, and it does not only broaden the scope of application of load balancing strategy, but also improve the operating efficiency of the load balancing.Third, A method for sever is presented to calculate dynamic load of server based on the characteristics of Computer system architecture, which evaluates a comprehensive load and dynamic load of server by analyzing the basic elements of affecting the sever load.Finally, by the network simulation tool, OPNET, this paper builds cluster models, designs load balancing module, and examines the content of above work. The results shows that the load-balancing strategy proposed based on the discrete uniform distribution has a good load balancing effect, and it can effectively reduce the workload of front-end balancer. Research findings in the paper are used in centralized cluster to relieve the work pressure of balancer and the possibility of its becoming the system bottleneck.