Cluster Load-Balancing Research Based On Content Classification And Session Sharing

Load-balancing has great significance in improving the effective performance of the Web cluster server to which the internet request explosion has brought huge pressure. The existing load-balancing system often leads to great request-delays and load-imbalance due to its large overhead and session-persisting. The load-balancing overhead can be reduced significantly through content classification, and the impact of session-persisting on load-balancing can be reduced by session sharing. The work done by this thesis is as follows:First, it conducts a profound study and comparison of the existing load-balancing algorithms. Then it explores the main contents of the load-balancing algorithm based on an in-depth analysis of the structure of the cluster. Besides, it provides a classification and comparison of the typical load-balancing algorithms in accordance with a variety of different optimization objectives and then points out the shortcomings of the existing load-balancing algorithms.Second, a content-based and dynamic load-balancing algorithm of the Web server is proposed in this thesis to solve the problem of excessive cost in the load-balancing for Web cluster. The algorithm encodes and classifies the contents, and then divides nodes into group and selects node according to the transfer request forwarded by the classified information. At a fixed interval of time the algorithm uses classification information to calculate the load of nodes, and forwards the request in accordance with the forwarding probability within the group. Simulation results show that under certain circumstances, the proposed algorithm in this thesis exhibits an advantage in performance improvement while compared with other algorithms.Third, a session sharing scheme for load-balancing is proposed in this thesis to solve the problem of the Web cluster load-imbalance caused by session-persisting. The scheme moves session information to a global-visible third-party session server from a single point, and then introduces cluster structure to the session server so as to guarantee its high availability. Comparative analysis shows that the proposed scheme is able to prevent the load-balancing schedule moving from packet level to stream level, thus avoiding the impact of session-persisting upon load-balancing; In addition, the scheme also builds conditions for a more sophisticated load-balancing algorithms and security policy.