Research On Dynamic Load Balancing Algorithm Based On LVS Cluster

Cluster technology is the technology that uses a certain topology to organize the computer and efficiently complete the specific task.In recent years,the Internet business has developed rapidly,and the high concurrency request of the user has brought unprecedented pressure on the server.The technology of load balancing cluster has become an important solution to the problem of high concurrency.The load balancing means that the load is evenly distributed to multiple server nodes in a cluster by using an effective network structure and using a reasonable scheduling algorithm.Linux Virtual Server is an excellent open source load balancing software in China.Because of its excellent performance and flexible way of use,it has become a popular load balancing solution.However,with the increase in the demand for Internet services,the load scheduling algorithm in LVS is gradually unable to meet the current requirements.In a continuous high concurrency environment,the load of the LVS cluster may be seriously unbalanced,leading to a decline in the quality of cluster services.So it is necessary to improve the performance of the LVS load scheduling algorithm to balance the cluster load.This thesis first gives a comprehensive introduction to cluster and load balancing technology.Then it analyzes the architecture of LVS,expounds the implementation principles of two modules of IPVS and KTCPVS in LVS,and compares the performance of four load balancing modes and ten load scheduling algorithms of LVS,and then summarizes the shortcomings of LVS's existing scheduling algorithm.On the basis of the above work,this thesis analyzes the principle of consistent hash algorithm,and then proposes a dynamic load balancing algorithm DCH to adapt to Web's high concurrency environment based on the advantage of consistent hash in load scheduling.The main design of the DCH algorithm is as follows: the load balancer periodically collects the state information of each server;the performance of the server CPU,memory and disk I/O can be used to quantify the performance of the server,and the shortcomings that the server performance evaluation depends on the administrator's subjective set of weights of the LVS built-in scheduling algorithm are improved;the utilization of the integrated server CPU,memory and disk I/O quantifies the load,improving the shortage of the LVS built-in scheduling algorithm to rely on the number of connections to measure the load;the number of virtual nodes is calculated using quantified performance and load,and it is used to measure the server's real-time processing request capability and periodically adjust the request allocation probability of each server,so as to achieve good load balancing effect.In addition,the implementation of the DCH algorithm in LVS is also described in this thesis.The following work is done in the experimental part of this thesis.First,the two important parameters of the DCH algorithm,the number of reference virtual nodes,the number of N and the periodic T,are determined by experiments.Then the request response time,maximum throughput and fault tolerance of the cluster are tested when the DCH algorithm is used,and compared with the original WRR and WLC algorithms of LVS.Experiments show that DCH algorithm shortens the response time of clusters,improves the maximum throughput of clusters,and has better load balancing ability and fault tolerance.