Research And Application Of Load Balance Technology Based On Web Server Cluster

Since the middle of twentieth century, Internet entered a period of rapid development, Web server is providing a variety of network services,this change promoted the number of Internet users and network traffic to grow exponentially, some similar online ticket reservation system,online buying system and other special site visit soaring in an instant. This gives the huge traffic pressure to popular website,a single Web server has been unable to meet the needs of the network.How to create a rapid response ability, high availability and easy management’s high-quality services to meet the growing load demand has become a pressing issue, which requires a high resistant to access and provide low-latency Internet infrastructure,Web server cluster system came into being, but how to choose the best server to answer the request, how to maximize the performance of the server cluster, a reasonable allocation of network request becomes a problem, therefore, the load balancing mechanism came into being.Load balancing can balance factors of conflict, such as cost, performance and scalability and implement strong performance that a stand-alone system can not complete through a relatively low total cost,so research of server load balancing technology has a very bright future. This paper presents a dynamic feedback load balancing algorithm,back-end real server collect their load information adaptively,then send to the load balancer, which reduces the communication overhead and the burden on the load balancer.In an effort to meet the increasing demand for Web applications and services, Web servers clusters by the usage of storage area network have been effective to satisfy the need.To reduce the communication overhead, these clusters rely on high-performance, low-latency intra-cluster communication mechanisms, such as the Virtual Interface Architecture(VIA) or InfiniBand Architecture,under these new architectures, many programmable NICs have come with large on-chip memory to support more connections,however, this abundant NIC memory space is not fully utilized.Server Clusters based on SAN,we have proposed two NIC caching schemes,namely,exclusive caching and inclusive caching,the exclusive caching scheme uses the NIC memory as an extended cache for the main memory cache,this scheme increases the cache hit ratio, and thus reduces the number of disk accesses,the exclusive scheme is particularly effective when the number of Web data items is large and the popularity skewness is low.The inclusive caching scheme maintains a copy of the most frequently accessed data items or files in the NIC memory,the main advantage of this scheme is to reduce the intra-cluster communication traffic and the DMA latency,the inclusive scheme is particularly effective when the size of Web content is large and the popularity skewness is high.