Study And Realization Of Dynamic Scheduling Based On Linux Virtual Server Cluster

With the development of Internet, web-based data visit grows rapidly. And the visits to data center, large enterprises, and web portals are up to the level of10Gb/s especially. Meanwhile, using HTTP、FTP、SMTP and other application program, the server websites provide abundant contents and information, and the server itself is submerged by large amount of data. Besides, many websites (especially those e-commerce websites) need to provide24/7service all the year around. Any interruption in the service or lose of key data would cause direct business loss. All the above requires the application service to provide high performance and high stability.However, comparing with the development of Internet technology, the growth of server’s processing speed and memory’s accessing speed are far slower than the growth of network bandwidth and the application service. The number of customers grows with the growth of network bandwidth, which consumes large amount of the server’s resources and makes the server the bottleneck of Internet.Linux Virtual Server (LVS), using load balancing technology, connect the servers into a virtual server cluster. It provides the fast growing Internet with an easy-expanded, practical and low price solution.The key technology in the cluster system is load balancing.The main function of load balancing is to transfer the service request to a real server according to a certain dispatching algorithm.Its purpose is to fully use every server in the cluster to shorten the service response time of the whole system.This paper does a profound research into existing dispatching algorithm of load balancing, analyzes the WLC algorithm, and finds out that the weight of load balancing dispatching is static and the dispatcher can’t adjust the weight dynamically according to the real load condition of the server, which could cause workload skew to the server.So, this paper proposes some improvement, introduces a dynamical feedback mechanism in adjusting weight in order to reflect the real load condition of the server cluster, and verifies it with experiments to prove that the improved algorithm is better than the existing algorithm to some extent.