Research On Cluster Load Balancing Algorithm Based On Dynamic Feedback Mechanism

With the development of the Internet, today's society has become information-oriented social and people set an even higher demand on the network servers. Memory, CPU and hard disk on traditional servers have substantial limitations, and only upgrading hardwares the servers are also difficult to adapt to current needs. The server-side is becoming the bottleneck in many applications. With the features of high reliability, high availability and scalability, the technology of server cluster is an effective way to achieve high-performance services, and can solve this problem.How to reasonable allocating tasks between the nodes in a cluster, and avoiding one node busy while the others can not play their best performances requires a reasonable load balancing algorithm to assign tasks among the nodes. However, the traditional load balancing algorithm has many flaws, such as it is not take the real-time load of nodes into account, so it can not balance the load in a cluster according to the real-time load of nodes. In this thesis the feedback mechanism is applied to the dynamic load balancing algorithm. In this way, it can receive the real-time load information of the node, and can keep the cluster at a balanced state, to reduce the average response time. The main work of this theis is as follows:(1) After an in depth study on Linux-based server cluster, with the actual situation, then proposed a load balancing strategy based on dynamic feedback mechanism combining static and dynamic. The algorithm considers the actual load and the processing performance of each server node, it can better reflect the real-time processing capability of each node, balance server load, and let each node can play their maximum processing performance. At the same time, dynamic feedback mechanism needs to continuely to collect information of each node, whlie will increase additional overhead. Considering the advantage of static algorithm is simple and with small cost, the proposed algorithm make a combination of static algorithm and dynamic algorithm, thereby improving the system throughout and reducing the average response time of overall system.(2) Base on the analysis of the dynamic feedback model and the popular cluster systems, a framework for implementing the algorithm is designed and realized under Linux. This system mainly performs the following functions. First, collects node information; then elects the available node according to the information and the improved load-balancing algorithm;at last the application complets the request on the election node and return the result. To show the running state of each node and the overall operation of the cluster, the system adds the real-time monitoring module.(3) Using the current popular performance testing tools—loadrunner, the comparison study between the current popular algorithm and the proposed one is carried out. The experimental results show that the proposed algorithm has a better balance results, reduces the response time of requests, improvs resource utilization of each server node.