The Algorithm For Load-balancing Of Web Clustering Dealing With The Mixed-resource Requests

With the rapid development of the Internet and its wide application,the business form of Web information system processing is various,and the services site scale and service response are increasing.In order to improve the availability,reliability and security of the information service system,the service cluster technology has emerged.Aiming at give full play to each node's resource and overall clustering performance,it is necessary to select the appropriate load-balancing technology and efficient load-balancing algorithm to allocate the client's requests so that basically balancing the size of each server or service node processing access load.Aiming at the insufficiency of the traditional load-balancing algorithm in Web clustering,this dissertation proposes a load-balancing optimization algorithm for processing mixed requests.The three types of improved models are mainly used to complete the following work:1.Load model optimization: the traditional load-balancing algorithm uses a single variable such as the number of connection and response time as the basis for measuring the load of the server.It does not consider other influencing factors and does a unified processing of the requests.The load model divides requests into static-resource requests and dynamic-resource requests.The multi-dimensional variables are used to calculate the node's maximum processing capacity,total load and static load.The model uses load rate to indicate the degree to which a node bears the load and proposes the concept of load-balancing variance to measure the load-balancing of the clustering.2.Self-adaptive weight model optimization: the traditional algorithm sets the server weight based on the server performance or the operator's experience.This weight is not accurate enough and cannot be updated.Because there is some non-linear relationship between the load of the server and the average time,the self-adaptive weight model uses BP neural network's error back propagation algorithm to store this nonlinear relationship so that the server weight can be corrected in real time according to the load of the cluster nodes.The experiment verifies that the model can effectively reduce the error between the output valueand the expected value so that each server can receive the expected dynamic load.3.Distribution model optimization: the traditional algorithm will allocate all requests to the same server within one sampling period,causing the load of the clustering to be skewed.The distribution model is based on an self-adaptive weight model,which uses the hashtable to record the allocation of static-resource requests and distributes the same request to the same server to improve the cache hit rate.This model uses the output error of the BP neural network to calculate the probability that each server receives the dynamic-resource requests,so that the model can dynamically distribute dynamic-resource requests within a sampling period to avoid load skew.The experiment verifies the validity of the model.4.This dissertation uses OPNET software to simulate the load-balancing algorithm.Compared with weighted round-robin algorithm and least-connection scheduling algorithm,experimental results show that the optimization algorithm minimizes the CPU utilization of load balancer and the load-balancing variance,achieves the best load balancing effect.Experiment proves the effectiveness and rationality of the optimization algorithm in this dissertation.