基于LVS集群的动态负载均衡策略的研究与实现

发布时间：2018-06-27 01:03

本文选题：集群技术 + Linux虚拟服务器　；参考：《北京邮电大学》2013年硕士论文

【摘要】：随着网络服务的不断发展,用户的访问量和服务器的负载量呈现指数级的增长,如何向用户提供高质量、不间断的网络服务,成为网络服务器供应商必须解决的难题。利用单一的服务器结构很难满足此需求,而集群技术的应用使得该问题迎刃而解。集群技术就是将一组相互独立的、通过高速网络互联的计算机构成一个整体,并通过单一系统的模式加以管理,使得用户与集群相互作用时,集群像是一个独立的服务器。作为集群实现技术的一种Linux虚拟服务器(LVS)已被广泛关注和应用。尽管目前LVS集群系统的应用较为广泛,但是LVS集群在负载均衡方面还不完善,现有的几种负载均衡算法并不能根据真实服务器的实时负载情况,来及时的调整负载的分配策略,这样很容易导致系统出现负载倾斜的现象,使集群性能大大降低。针对此问题,本论文提出了一种能够动态调整服务器权值的动态负载均衡策略,该策略能够有效地避免集群出现负载倾斜的现象,进而提升集群整体的性能。论文主要涉及了以下工作：首先,研究了集群系统和负载均衡技术,对两者有了初步的了解。然后,深入研究LVS集群技术,对其体系结构、工作原理和三种工作模式有了较为深刻的认识。同时还研究了现有的八种负载调度算法,并重点对WLC算法进行了深入分析,总结其存在的不足之处。接着,针对WLC算法存在的缺陷,提出了动态负载均衡策略的思想,设计并实现了相应的动态负载模块(Dynamic-Load Module)。再次,搭建简单测试环境,利用压力测试工具---Web Application Stress Tool(WAS),对原有的WLC算法和改进后的动态负载均衡策略进行性能测试,通过对实验结果的对比分析,初步验证了动态负载均衡策略能够有效地提升集群系统的性能。最后,利用中国移动研究院的PDM-PAAS项目的部署环境,对动态负载均衡策略进行复杂环境下的性能测试,也采用了WAS压力.测试工具,对原有的WLC算法和动态负载均衡策略进行了性能对比测试,并对实验数据进行了对比分析,进一步验证了动态负载均衡策略的有效性,并能够将其应用在PDM-PAAS项目中。
[Abstract]:With the continuous development of network services, the number of users and the load of servers increase exponentially. How to provide users with high quality and continuous network services has become a difficult problem that network server providers must solve. It is difficult to satisfy this requirement by using a single server structure, but the application of cluster technology can solve the problem easily. Cluster technology is a group of independent computers interconnected by high-speed networks to form a whole, and through a single system model to be managed, so that when users interact with the cluster, the cluster is like an independent server. As a cluster implementation technology, Linux Virtual Server (LVS) has been widely concerned and applied. Although the LVS cluster system is widely used at present, the load balancing of LVS cluster is not perfect. The existing load balancing algorithms can not adjust the load allocation strategy according to the real server's real-time load situation. This can easily lead to the phenomenon of system load tilt, so that the performance of the cluster is greatly reduced. To solve this problem, this paper proposes a dynamic load balancing strategy which can dynamically adjust the weights of the server. This strategy can effectively avoid the phenomenon of load tilt in the cluster and improve the overall performance of the cluster. The main contents of this paper are as follows: firstly, the cluster system and load balancing technology are studied. Then, the LVS cluster technology is deeply studied, and its architecture, working principle and three working modes are deeply understood. At the same time, eight existing load scheduling algorithms are studied, and the WLC algorithm is analyzed in detail, and the shortcomings of WLC algorithm are summarized. Then, aiming at the shortcomings of WLC algorithm, the idea of dynamic load balancing strategy is put forward, and the corresponding dynamic load module is designed and implemented. Thirdly, a simple test environment is built, and the original Application stress tool (was) is used to test the performance of the original WLC algorithm and the improved dynamic load balancing strategy. It is preliminarily verified that dynamic load balancing strategy can effectively improve the performance of cluster system. Finally, the deployment environment of PDM-PAAS project of China Mobile Research Institute is used to test the performance of dynamic load balancing strategy in complex environment, and was pressure is also used. The performance of the original WLC algorithm and the dynamic load balancing strategy is compared with the test tool, and the experimental data are compared and analyzed. The validity of the dynamic load balancing strategy is further verified, and it can be applied to PDM-PAAS project.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.5

【参考文献】