基于RPC的高并发网络通信中负载均衡的研究

发布时间：2018-05-23 15:21

本文选题：远程过程调用 + 高并发　；参考：《浙江理工大学》2017年硕士论文

【摘要】：随着互联网相关技术的快速发展,数以万计甚至是上亿的用户在不断地访问大型网站应用,它们的访问数量呈指数级别增长,特别是活动时期的访问数量更为庞大。高并发指多个用户同时发送请求,此情况下服务器资源利用率相应上升,到达阈值后服务器性能会缓慢下降甚至骤然降低。面对如此海量的并发请求时,后台服务要保持原有即时、流畅、稳定的用户体验,就需要有良好的网络通信架构的支撑,本文选择对RPC进行深入研究。本文首先分析高并发的特点,对请求的响应时间以及服务器集群的动态硬件资源监控进行了研究和分析,提出了一种基于RPC的动态负载均衡策略,设计相应的数学模型,实现多台服务器集群系统进行并发测试,通过与常用RPC框架的对比,验证该策略的可行性以及有效性。提出一种针对高并发的评价模型,进行结果分析。最后,为了更好地满足高并发的需求,调整模型的可变参数进行多次实验,拟合函数求最优解实现模型优化。本文主要工作有以下几个方面:(1)对常用负载均衡以及常用RPC框架进行研究,理解其工作原理。对其中的Dubbo框架的优势和自带负载均衡进行研究,并对此四种自带负载均衡进行实验测试,将数据汇总进行对比,选择更适合高并发的轮询负载均衡进行下一步的研究。针对RPC的请求调用特点,提出RPC请求的响应时间模型,结合最快响应算法以及硬件资源监控,提出了动态权重值更新策略以及相应的数学模型;(2)根据本文提出的动态权重值更新策略,提出了基于RPC的动态加权轮询负载均衡的策略,并进行核心模块的划分与设计,包括动态注册模块、数据缓存模块、核心监控处理模块以及负载均衡模块,并介绍了各个模块之间的关联;(3)对已划分的模块进行实现,搭建出三台服务提供方的集群。通过与常用RPC原生框架的实验对比,验证了本策略的可行性,它既能减少平均响应时间,在高并发的场景下已经超过常用的RPC框架的效率,也可以充分地利用性能较好的服务提供方;(4)最后,提出了一种针对负载均衡效率的评价模型,用于模型参数优化。通过可变参数的调整获得不同条件下的负载均衡性能对比,通过实验结果拟合出一个针对本研究的二次拟合函数,经过法方程求解,求出本实验的最优解,并给出理想化的参数值。
[Abstract]:With the rapid development of Internet related technologies, tens of thousands or even hundreds of millions of users are constantly accessing large web applications. Their number of visits has increased exponentially, especially during the activity period. When the threshold is reached, the performance of the server will decline slowly or even suddenly. In the face of such massive concurrent requests, the background service needs to be supported by a good network communication architecture to maintain the original, fluent and stable user experience. This paper chooses the RPC to study deeply. This paper first analyzes the characteristics of high concurrency, and requests the request of the high concurrency. The response time and the dynamic hardware resource monitoring of the server cluster are studied and analyzed. A dynamic load balancing strategy based on RPC is proposed, and the corresponding mathematical model is designed to realize concurrent testing of multiple server cluster systems. By comparing with the common RPC framework, the feasibility and effectiveness of the strategy are verified. A high concurrency evaluation model is proposed for the analysis of the results. Finally, in order to meet the needs of high concurrency better, the variable parameters of the model are adjusted for many experiments and the fitting function is used to optimize the model. The main work of this paper is as follows: (1) research on the common load balancing and the common RPC framework. The advantages of the Dubbo framework and the load balancing are studied, and the four kinds of load balancing are tested, the data collection is compared, and the next step of the polling load balancing is selected more suitable for high concurrency. The response time mode of the request of the RPC is put forward according to the request calling features of the RPC. Type, combined with the fastest response algorithm and hardware resource monitoring, the dynamic weight updating strategy and the corresponding mathematical model are proposed. (2) according to the dynamic weight updating strategy proposed in this paper, a dynamic weighted polling load balancing strategy based on RPC is proposed, and the core module is divided and designed, including the dynamic registration module. According to the cache module, the core monitoring and processing module and the load balancing module, the relationship between each module is introduced. (3) the divided module is implemented and the cluster of three service providers is set up. The feasibility of this strategy is verified by comparing with the common RPC primary framework, which can reduce the average response time and be high. The concurrent scenario has exceeded the efficiency of the common RPC framework and can make full use of the service provider with better performance. (4) finally, an evaluation model for load balancing efficiency is proposed for model parameter optimization. The performance comparison of load balancing under different conditions is obtained through the adjustment of variable parameters, and the experimental results are obtained. Fitting a two fitting function for this study, through the solution of the equation, the optimal solution of the experiment is obtained, and the idealized parameter value is given.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【相似文献】