The Design And Implementation Of IP Control Gateway Cluster Based On Time Series Prediction

In order to resolve the issue that the bandwidth of IP control gateway (IPCG) isn't enough to forward packets, this thesis proposes the IPCG cluster based on time series prediction, implements the load balancing algorithm which is based on time series prediction and the technique of load balancing which is based on Netfilter and policy routing.Firstly, the thesis introduces the concept of cluster, the concept of load balance cluster and the most important factors in load balancing, such as scheduling algorithms, network topology and scheduling granularity. Secondly, it introduces the indexes of the server load, the computing methods and algorithms, and analyzes the disadvantages of algorithm. Combined with the changing regularity of the IPCG's load, it proposes the prediction algorithm of load balancing based on time series, and expounds the process of building the model of time series according to using the IPCG's historical load data in details. It also applies the ARMA(p, q) model to predict the load of IPCG. The model is proved to be more efficient in the process of prediction. Thirdly, it introduces the general three load balancing techniques which are NAT, IP tunneling and DR. On the base of analyzing the framework of Netfilter and policy routing, it proposes the half-connection state detection method and the load balancing technique which fit to IPCG cluster. Finally, it compares the performance of the algorithm based on time series prediction with the algorithm of least-connection scheduling in the IPCG cluster and compares the network throughput of IPCG with IPCG cluster, then makes out the conclusion that the algorithm and the implementation method presented in the thesis can not only get a high bandwidth, but also gain the effect of load balancing in IPCG cluster. The algorithm and the technique can be used in the new generation network systems which link the Intranet to the Internet, such as the firewall cluster, the IDS cluster, IPCG cluster and so on.