基于JCF中间件的负载均衡算法研究

发布时间：2018-06-14 16:40

本文选题：JCF中间件 + 负载均衡　；参考：《中国民航大学》2015年硕士论文

【摘要】：随着中国民航旅客逐渐增长,旅客网络订票数量也呈爆发式增长,对处理订票请求的分布式服务交易平台——JCF(Java Component Framework)系统造成了很大压力。目前部署在JCF中间件上的负载均衡算法是静态加权轮询算法,该算法在提高系统吞吐量及降低响应时间方面效果不显著,易造成负载不均,需要改进当前的负载均衡算法。首先,本文调研了中间件和负载均衡算法的国内外现状,着重分析了JCF中间件的特点以及其负载均衡机制,并研究目前部署在其上的静态加权轮询算法,为后续的算法改进做好了基础工作。其次,分析了静态加权轮询算法存在的不足,提出了一种周期可变的动态负载均衡算法。该算法将反馈负载信息的周期长度与网络中请求数量相关联,既保证了反馈信息的实时性又有效地降低了计算各节点负载量的资源消耗。在该算法实现过程中,采用改进的模拟退火算法确定负载权重向量β,具体思想是将响应时间作为最优解,选取得到最优解的那组β值,提高了人为经验得到β值的精确性。最后,提出了一种负载预测与过载迁移的融合算法,采用对周期内无法提取到的负载情况进行预测,对超过负载阈值的服务器进行告警,并且关闭该进程将服务请求分发给轻负载服务器的方式。通过提前预测与负载迁移的有机结合,解决了周期内无法获取各个节点的实际负载问题,负载预测的不准确问题,以及可能出现的部分服务器过载,甚至崩溃问题,最终达到了系统的负载均衡。在中间件平台上的实验结果表明该算法优于静态加权轮询算法,动态加权轮询算法以及周期可变的动态负载均衡算法,在提高均衡效率,增加集群系统的吞吐量,降低服务请求响应时间等多方面有着显著效果,在实际应用中有较大的价值。
[Abstract]:With the gradual growth of civil aviation passengers in China, the number of tickets booked on the passenger network is also increasing explosively, which has caused great pressure on the JCFN Java component Framework system, which is a distributed service trading platform for handling booking requests. At present, the load balancing algorithm deployed on JCF middleware is static weighted polling algorithm. This algorithm is not effective in improving system throughput and reducing response time, it is easy to cause load imbalance, so it is necessary to improve the current load balancing algorithm. Firstly, this paper investigates the current situation of middleware and load balancing algorithms at home and abroad, analyzes the characteristics of JCF middleware and its load balancing mechanism, and studies the static weighted polling algorithm deployed on JCF middleware. The basic work is done for the subsequent algorithm improvement. Secondly, the shortcomings of static weighted polling algorithm are analyzed, and a dynamic load balancing algorithm with variable cycle is proposed. The algorithm associates the cycle length of the feedback load information with the number of requests in the network, which not only ensures the real-time performance of the feedback information, but also effectively reduces the resource consumption for computing the load of each node. In the implementation of the algorithm, the improved simulated annealing algorithm is used to determine the load weight vector 尾. The concrete idea is that the response time is taken as the optimal solution, and the group of 尾 values obtained from the optimal solution is selected, which improves the accuracy of the artificial experience to obtain the 尾 value. Finally, a fusion algorithm of load prediction and overload migration is proposed. And shut down the way the process distributes service requests to light-load servers. Through the organic combination of early prediction and load migration, the problems of the actual load of each node can not be obtained in the cycle, the inaccuracy of load prediction, and the possible overload or even crash of some servers are solved. Finally, the load balance of the system is achieved. The experimental results on the middleware platform show that the proposed algorithm is superior to the static weighted polling algorithm, the dynamic weighted polling algorithm and the dynamic load balancing algorithm with variable cycle, which can improve the balancing efficiency and increase the throughput of the cluster system. Reducing the response time of service request has remarkable effect and has great value in practical application.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V354

【参考文献】