基于自适应小生境遗传算法的LVS负载均衡调度研究

发布时间：2018-04-29 20:38

本文选题：自适应小生境遗传算法 + 负载均衡　；参考：《西南交通大学》2013年硕士论文

【摘要】：随着Internet的发展,网络用户日益增加,需要服务器系统具有处理高并发访问的能力。目前,服务器集群是解决服务器超载问题最有效的解决办法。它将多个服务器集中起来并行工作,配合共同完成同一种服务；对客户端而言,相当于只有一台超级服务器在进行服务。其工作模式不需要更改任何服务器端和客户端代码,可靠性强。目前服务器集群中的调度算法由于过于简单,存在扩展性较差和并行性较弱等问题,为了改善和提高集群系统性能,本文从算法和架构上对Linux虚拟服务器(Linux Virtual Server, LVS)进行了研究。首先在分析遗传算法理论的基础上研究其在调度作业中的应用,并根据服务器负载调度的需求,选择自适应小生境遗传算法(Adaptive Niche Genetic Algorithm, ANGA)作为服务器集群的调度算法,提出了相对CPU利用率作为适应值的评价标准,并合理应用于负载均衡调度作业中；然后分析了LVS的工作原理及其缺点,对常规的服务器集群架构进行了改进,提出基于决策器训练调度序列；进而设计并实现了以下四个关键模块单元,即调度算法单元、负载均衡器(Load Balancer,LB)与决策器的通信单元、决策器与节点间的通信单元以及修改后的IP虚拟服务器(IP Virtual Server, EPVS)内核单元,并对通信协议以及工作流程做了详细的介绍。最后,通过搭建真实服务器集群平台,在LVS-NAT模式下测试了改进后的负载均衡调度系统,实验利用JMeter等压力测试软件模拟大并发,并利用Wireshark对网络数据包的流向进行监控。实验结果表明,本方案的系统平均响应时间仅为2毫秒,同时错误率趋近于0,相比改进前服务器系统性能更稳定。
[Abstract]:With the development of Internet, network users are increasing day by day, which requires server system to handle high concurrent access. At present, server cluster is the most effective solution to the problem of server overload. It gathers multiple servers to work in parallel, working together to complete the same service; to the client, it is equivalent to only one super server serving. Its working mode does not need to change any server-side and client-side code, strong reliability. In order to improve and improve the performance of Linux virtual server system, this paper studies the algorithm and architecture of Linux virtual server, in order to improve the performance of cluster system, because the scheduling algorithm in server cluster is too simple, there are some problems such as poor expansibility and weak parallelism. Firstly, based on the analysis of genetic algorithm theory, the application of genetic algorithm in scheduling job is studied, and according to the demand of server load scheduling, adaptive Niche Genetic Algorithm, ANGA) is selected as the scheduling algorithm of server cluster. This paper puts forward the relative CPU utilization rate as the evaluation criterion of fitness value, and applies it to the load balancing scheduling job reasonably, then analyzes the working principle and shortcomings of LVS, and improves the conventional server cluster architecture. The following four key module units are designed and implemented: scheduling algorithm unit, load balancer load Balancer-LBB) and Decision-Making unit. The communication unit between the decision maker and the node and the modified IP virtual server IP Virtual Server (EPVs) kernel unit are introduced in detail. The communication protocol and workflow are introduced in detail. Finally, by building a real server cluster platform, the improved load balancing scheduling system is tested in LVS-NAT mode. The pressure test software such as JMeter is used to simulate the large concurrency, and Wireshark is used to monitor the flow of network data packets. The experimental results show that the average response time of the system is only 2 milliseconds, and the error rate approaches zero, which is more stable than that before the improvement.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.5

【参考文献】