| With the development of Internet technology, more and more traditional server are replaced by distributed system. Under this background, load balance algorithm is becoming a more and more popular topic. How to take advantage of every single node in a cluster, and how to ensure the quality of service is the goal of algorithm optimization.First, we give a brief introduction about the application background of RPC and the reason to replacement traditional procedure call with it. In the environment of Web2.0, distribution technology is a trend and the key point of modularization development and quality of service. Then we talk about the history of distributed system and some existed load balance algorithm.In the next part, we introduce the background of LVS system and the eight build-in static schedule algorithms, especially the weighted least connection, which is the default algorithm of LVS system. Because of its the disadvantage like inflexibility and cannot deal with the mutation of flow rate, we propose a new algorithm based on the feedback. The new algorithm collect servers’CPU load, memory usage and10situation, calculate the load value of a specific server, then adjust the weight of e every node dynamically. The new algorithm has a higher complexity, but can make up some drawback of the default one. Then, we give the implementation and design of our algorithm, including the weight calculation module of load balance computer and status gather module in servers.In the last part, we build up a small cluster, use a Microsoft’s free software called web application stress tool to simulate multiple users and run some concurrent stress tests. Then we compare the internal properties by assign two identical servers with different initial values. Results show that the improved algorithm has lower response time and higher capacity. In the second experiment, the two nodes’ weight will get closer after a period of time, so the new algorithm can effectively avoid load imbalance. |
PDF Full Text Request