流媒体服务器集群负载均衡策略的研究

发布时间：2018-04-04 07:27

本文选题：流媒体　切入点：集群　出处：《西安邮电大学》2017年硕士论文

【摘要】：随着互联网及计算机技术的快速发展,如今4G网络已经普及,人们对流媒体业务的需求越来越高,单个流媒体服务器已不能满足用户的需求,进而采用流媒体服务器集群来解决。负载均衡作为集群的一种核心技术,能够提高集群的系统性能,其效果通常取决于负载均衡算法的设计。传统的动态反馈算法虽然能够定期的获取集群节点的负载状况,但是当周期内某集群节点的负载变化较大时,就不能及时的反馈其真实的负载状况,而且不能较好的处理突发的大量并发连接请求,有可能引发集群的负载倾斜现象。本文首先介绍了集群负载均衡技术的发展以及流媒体的相关协议。其次,深入的分析与研究了现有的静态和动态负载均衡算法,明确各种调度算法的适用场景。然后,针对传统动态反馈算法存在的问题做出如下改进:1、根据集群节点每秒钟任务连接数的变化量动态的修改负载反馈周期,提升集群节点负载反馈的及时性。2、将集群节点按其负载状况分为低负载、正常负载、高负载三类,类之间根据总的负载权值进行任务分配,类中采用最小连接数算法分配任务,解决处理大量并发任务请求时存在的负载倾斜问题。3、采用流媒体服务器的中继/转发功能实现过载节点的负载迁移,进一步提升集群的负载均衡效果。最后,本文对优化的负载均衡调度策略进行编码实现,并搭建流媒体服务器集群,对优化的调度策略进行验证,并同传统的动态反馈负载均衡算法和最小连接数算法进行比较。实验结果表明,优化的调度策略能够更加及时的反馈集群节点的负载状况,能有效的解决大量并发请求导致的集群负载倾斜问题,提升集群的负载均衡效果和集群的服务质量。
[Abstract]:With the rapid development of Internet and computer technology, now 4G network has spread, people demand for streaming media service is more and more high, a single streaming media server has been unable to meet the needs of users, and then used to solve the streaming media server cluster load balancing. As a core technology of the cluster, can improve the performance of the system design of the cluster. The effect usually depends on the load balancing algorithm. The traditional dynamic feedback algorithm can obtain the cluster nodes load status regularly, but the load changes greatly when a cluster node cycle, can not be timely feedback on the real load conditions, but can not handle a large number of concurrent unexpected connection requests may lead the cluster load tilt phenomenon. This paper firstly introduces the related protocol development of load balancing technology and streaming media. Secondly, deeply The analysis and Research on the existing static and dynamic load balancing algorithm, clear application scenarios for each scheduling algorithm. Then, based on the traditional dynamic feedback algorithm problems make the following improvements: 1, according to the cluster node dynamic changes of the number of connections per second task modified load feedback cycle, enhance the cluster node load feedback timeliness.2 the cluster node, according to the load condition for low load, normal load, high load three class, between classes according to the total weight of load assignment, task allocation algorithm using the minimum number of connection class, to solve the processing of the existence of a large number of concurrent tasks at the request of the skew.3, the relay / streaming media server forward function to transfer the load overload node, to further enhance the effect of load balance cluster. Finally, the load balancing scheduling strategy to optimize the coding, and so on The construction of streaming media server cluster, the optimized scheduling strategy is verified and compared with the traditional dynamic feedback load balancing algorithm and the minimum number of connection algorithm. Experimental results show that the optimal scheduling strategy can feedback the cluster nodes load more timely, can effectively solve the large number of concurrent requests to the cluster load tilt problem to enhance the effect of load balance and cluster quality of service clusters.

【学位授予单位】：西安邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.05

【参考文献】