Research On Load Balancing Of Web Servers For Wan

With the massive application of Web services and streaming media,the amount of concurrency of Web sites has become larger and larger.In response to high concurrency requirements,multiple physical areas deploy Web clusters and CDN acceleration are current solutions.Therefore,how quickly the cluster's load balancer can obtain the information of the client node and use this information to locate the Web cluster or CDN resources that the node needs to access becomes more important.At the same time,the power consumption of the data center is very high.If the data center is used to carry the Web cluster in multiple physical locations,the energy consumption of the entire site will be higher.Moreover,most Web load balancing solutions only consider the user experience,do not consider the energy consumption of the cluster,and do not consider the operational benefits of the cluster.The characteristics of load balancing are also not considered when studying the energy consumption of the cluster,so that the two are always studied separately.In this paper,the primary job is as follows:The problem of cluster location for client nodes need to access,this paper proposes a new solution to location problem that cosine similarity location method based on DNS collaboration.This method use local DNS server and site DNS server(load balancer)to determine the location of request source.This solution not only has high availability,stability and scalability,but also reduces response time delay and improves the user experience.The problem of energy consumption for clusters,this paper argues that the best load balancing algorithm can't just consider one aspect of user experience.the best load balancing algorithm is not to distribute all tasks equally to each node,it should consider both the user experience and cluster energy consumption,other nodes are assigned tasks after a node reaches full load with an acceptable maximum processing delay.This not only effectively utilizes resources,but also reduces cluster energy consumption,finding a balance between cluster energy consumption and time delay.Therefore,based on load balancing,this paper proposes a scheme that based on prediction to dynamically reduce cluster energy consumption,the principle is to predict the number of service nodes required according to the current network flow,and then let the cluster dynamically control the number of nodes that need to be turned on or off,so that the requests and energy consumption ratio is always at a higher position.This solution can save a part of the no-load power and effectively reduce the energy consumption of the cluster.