| As a solution of mass storage, the NAS network storage has been widely used. With the continuous growth of the amount of data, a single NAS server can’t meet the demand of customer growth no longer. Therefore, the cluster server and the load balancing algorithm has become a hot topic in recent years. This article analyses the existing problems of load balancer algorithm at present that the Load Balancer itself will become the new bottlenecks when the data is very large. Besides with no consideration of the difference between the server performance and the differences between the size and computational complexity of task, the Load is not balanced very obvious. In order to solve the new bottleneck of the Load Balancer and improve the network throughput rate further, this article expands the TCP message and improves the TCP communication protocol that add new option information to keep client and server true IP address and port, then separates the Load Balancer from the internet and let the client connect the server directly. As a result, the Load Balancer has solved the new bottleneck and improved the network throughput rate. Secondly, this article puts forward an algorithm which is the weighted response time to reduce the response time of task that comprehensively consider the difference of the server and task. This article picks up k tasks which have been disposed by the server and dynamic predicts the response time of the new task based on the response time of the K tasks. Through the introduction of compensation factorε, it can improve the precision further. Besides, with the above of two improved algorithm, this article designed a Load Balancer and described the hardware and software model of the Load Balancer which is based on ARM9. The experimental results show that the load balancer reduces the average job of response time significantly. This system has low cost, strong extensibility and load capacity. |
PDF Full Text Request