Research On Technology Of Dynamic Load Balancing For CDN

With the growing of Internet users’ scale and the expanding of Internet services,content acceleration technology has been widely used. As one of the most effectivetechnique in content acceleration technologies, CDN(Content Delivery Network)has been applied to every aspect of Internet services, such as audio and videoservices, e-commerce and social network etc. CDN technology improves networkquality of service by deploying servers at the network edge, shortenging the accessdistance between the user and the website. How to balance the load of servers in thecontent delivery network is a main research diffculity to be considered.Firstly, this thesis analyzes the deployment level of existing CDN loadbalancers and load balancing algorithms. Considering the load pressure of serviceside, we use the load balancing mechanism based on network layer. Secondly, afteranalysis of practical Internet services, we find that the influence of common networkservices to the server’s load mainly include five aspects: CPU utility, memory usage,connection numbers, bandwidth, disk I/O number. The current weighted least-connection algorithm does not consider the influence of different applications’connection to the server load. This thesis uses above five metrics as the parametersof load balancing calculation, and proposes the variable factor weighted least-connection algorithm. The algorithm considers the different applications’ impacts onserver’s load and the server’s performance. It can adjust the load weight of servicesand reach a better performance in load balance. Frequent changes to server’s loadweight can result in concussion of server load scheduling, thus we segment therange of variable factor to decrease its influence on balance strategy. Then wedesign and implement a CDN load balancing system which based on LVS(LinuxVirtual Server). And we applied our variable factor weighted least-connectionalgorithm in this system. Finally, we compare our algorithm with typical weightedleast-connection algorithm in a real network experiment. Experiments show theeffectiveness of our variable factor weighted least-connection algorithm.