Load Balancing Of Virtual Machine Cluster

The arrival of Internet era pushes the conversion of network computing mode from parallel computing, distributed computing and grid computing to cloud computing. In front of cloud computing era, virtualization technique seems to be an efficient assistant, which equip cloud computing with characteristics such as more flexibility, scalability, lower cost, green and power-saving as well as dynamic deployment. And cloud computing belongs to the category of service computing, which in fact exists as a service based on web. The goal is to make users pay by demand while reducing the investment of traditional hardware, software and specialized technique.Facing with large numbers of users using cloud computing environment, load balancing is of great significance. Load balancer makes it possible for high-quality services and resource distributed well-proportioned, especially for machines with different hardware performance, the resources can be made the best of while under the maximum load that may lead to system halting.Traditional load balancing methods include LVS and DNS robin. LVS is short for Linux Virtual Server, which uses many servers to construct a virtual server cluster by load balancing technique. LVS provides a scalable, high-available and cost-saving solution to satisfy the increased network access requirement. However, it is a static load balancing open-source software running on the Linux operating systems and many famous load balancer producers embed it into hardware to implement static round robin load balancing function. In view of this fact, many domestic institutions have done some research about its dynamic feature, which is implemented by constant response from servers and modify their weight. So the dynamic weight can be changed with the variation of the server performance, effectively balancing servers.The emergence of cloud computing breaks traditional methods of load balancing. This thesis puts forward that load balancing is to serve cloud computing while providing services by implementing a more flexible load balancing system based on scalable and real-time responding characteristics.Because there is no standard definition about server load, this thesis defines load by pulsing and summing weights of server resources including CPU usage, RAM usage, and performance parameters such as response time, connection numbers and process numbers, concluding the best combination from a large number of experiment data. This thesis also discusses the best coefficients in different response time under different algorithms so as to probe the definition of load in the form of formula. Finally, a new method to construct a scalable cluster is put forward with virtualization technique, which has widely application scenarios and sufficiently shows the importance of the elasticity of the network service in cloud computing era.