A New Load Balancing Algorithm Of The Server Cluster

With the rapid development of Internet technology, the Internet has become theessential tools of more and more people to learn, work, and life. However, with theincreasing number of Internet users, the server systems of the major Internet companiesbecome more and more overwhelmed. With the cluster technology to retrofit theoriginal server system it must be a good method that can quickly deal with the situationand effectively solve this problem.This paper first studied the development status of the server cluster and loadbalancing technology, and then specialized in the design principles and softwareimplementation of the core of the Linux Virtual Server cluster system. Cluster systemsgenerally rely on the front load balancing scheduler to forward the user’s connectionrequests to a specific server on the background. Forwarding process must work with theload balancing algorithm, in order to allow users connection requests to get a reasonabledistribution between the servers. After the studies of the existing load balancingalgorithms in the Linux Virtual Server cluster system, this paper pointed out that thesealgorithms have these or those drawbacks, and the important one was that thesealgorithms did not base on the overall performance level and the real-time loadcondition of the servers in the background of the cluster system to reasonably allocatethe user’s connection request, and can not let the entire cluster system to achieve a stableload balancing state.After the in-depth studies on the Linux Virtual Server cluster system and its mainfour kinds of load balancing algorithms，the paper presents a new algorithm, namely,PCW load balancing scheduling algorithm. The most prominent feature of the PCWload balancing scheduling algorithm is that it is able to collect the performance andreal-time load indexes of each server in the background. The performance indexesinclude the CPU processing speed, memory size, disk read-write speed and the networkbandwidth. The load indexes corresponding to the four performance indexes are CPUutilization, memory utilization, the size of the average disk read-write rate and theaverage rate of the network. With the performance indexes the PCW load balancingscheduling algorithm calculates the specific server’s performance parameters; with theload indexes it calculates the specific server’s integrated load parameters. Through theanother formula in the paper and the two parameters the load balancer can gain the weight value of the specific server. Thus，by this method, in the running process of thecluster system the load balancer can access to the weight value which represents thespecific server performance levels and its real-time load, so that it can be morereasonable to allocate the specific users connection request to the servers.Finally, this paper set up a practical Linux virtual server cluster system testenvironment to test PCW load balancing scheduling algorithm and to compare with theexisting algorithm. The test results showed that PCW load balancing schedulingalgorithm obtained a better load balancing effect, and had great application value.