Research On JCF Middleware Load Balancing Algorithm

With the gradual growth of the China civil aviation passengers, the number of booking tickets also showed explosive growth online, which caused a lot of pressure to Java Component Framework(JCF) system. The JCF system is a distributed service trading platform, which is used to handle process reservation requests. Currently, the load balancing algorithm deployed in JCF middleware is the Static weighted round robin algorithm, which can not be good to improve system throughput and reduce response time, and could easily lead to load imbalance. So the load balancing algorithm needs to be improved.Firstly, this paper investigated the status of the middleware and the load balancing algorithms at home and abroad, analyzed the characteristics and the load balancing mechanism of the JCF middleware, and studied the static WRR algorithm currently deployed on it, which could help to improve the algorithm.Secondly, this paper analyzed the disadvantages of the static WRR algorithm, and proposed a variable period of dynamic load balancing algorithm. The algorithm makes the cycle length of feedback load information vary with the number of requests through network, and the system will not only automatically shorten the feedback cycle to ensure the real-time information feedback but automatically increase the feedback cycle to reduce resource consumption for calculating load of these nodes effectively. An improved simulated annealing algorithm is adopted to determine weight vector β. The specific idea is that the response time is as the optimal solution, and selects β in the optimal solution, which avoids the uncertainty through empirically obtaining value.Finally, this paper proposed a load forecasting and overload migration fusion algorithm. The algorithm can forecast the load for the quest of not obtaining the actual load of each node in a cycle interval, give the warning the server that it exceeds the threshold, close the processes, and distribute the service requests to the load on the server lightly. By combining forecasting and migrating load, the algorithm solved the questions that not obtained the actual load, adopted the inaccuracies of the load prediction, and avoided the overload even collapse situation of the part of the servers in the process of the algorithm. It ultimately achieved load balancing system. The results of experiments, which the variable period of dynamic load balancing algorithm, dynamic weighted round-robin algorithm and static weighted round-robin algorithm, are done in the JCF(Java Component Framework) middleware platform show that the algorithm in this paper has a significant effect in the respects of improving equilibrium efficiency, increasing throughput, and reducing service request response time about cluster system, and the algorithm has greater value in real applications.