Research And Improvement Of The Load Balancing Algorithm Based On The Lvs Cluster

With the growth of Internet, the pressure of Web server to handle the increasing user's requests is becoming bigger and bigger. In this situation, the way of replacing server's hardware is far from being able to meet the demand and expensive, also the performance is improved limitedly.The cluster can be a very good solution to deal with this problem. Cluster connects numbers of machines as one united computing resource to provide service for the external users. Appropriate size of cluster has strong computing power, and the price is much lower than a supercomputer. In addition, cluster also has good scalability and high reliability.The load balancing algorithm has great impact on the performance of cluster. Load balancing algorithm dispatches user's requests to each machine as evenly as possible, in order that the cluster can shorten the response time and improve the capacity of requests.This paper adopts the Linux virtual server (LVS) to construct the Web server cluster. After analyzing the LVS open source software and the existed load-balancing algorithm, the author proposes a new method to improve the performance of the cluster.The research works of this paper are as follows:1. Concludes and summarizes the current research situation of the cluster technology and load balancing technology.2. Describes the architecture of LVS and three ways of IP forwarding in load balancing, and then analyzes the composition, working mechanism and modules of IPVS which is the key part of LVS.3. Introduces Dynamic feedback load-balancing algorithm which can modify the real server'weight by collecting load information's, so that can avoid the shortcomings of the existed load-balancing algorithm,4. Constructs a LVS cluster with LVS-DR scheme, and uses Microsoft Web Application Stress Tool to make a simulation of huge numbers of users to test the Web server cluster. Results show that Dynamic feedback load balancing algorithm reduces response time from the server, and enhances the performance of the cluster.