基于负载预测的共享资源网络服务器节能控制研究

发布时间：2019-03-28 11:25

【摘要】：近年来，随着实时性应用的大量出现，，人们对服务质量的要求也随之提高。然而，人们在享受大规模集群服务器为人们提供的便利时，却忽视了随之而来的严峻问题----能耗。随着我国电荒严重程度的加深，加之国际经济危机对企业运营成本的管理提出了更高要求，服务器利用率的提高与节能已经成为了学术界和工业界关注的热点。本文的研究对于提高共享网络资源利用率和节能控制具有重要的意义。目前网络应用中，流媒体应用占较大份额，因此本文针对流媒体服务器日益增加的能耗问题，做了如下几个方面的工作： 1.在Linux系统下，本文针对许多文献尚未提及的硬盘存取能力进行了分析，改进了基于/proc文件系统和加载内核模块LKM相结合的方法，实现了对计算资源、内存资源、硬盘资源、网络带宽资源的测量。该方法利用/proc文件系统获取计算机信息全面、迅捷、准确的特点以及加载内核模块LKM占用资源小、访问权限高的优点，使之更能快速、准确、全面地获取了系统实时负载情况。 2.提出了一种加权负载求导预测法，实现了对流媒体集群服务器的网络资源的预测。该方法基于流媒体负载的日周期性，利用负载曲线变化率的稳定性进行预测工作。对现有加权平均计算方法中只是针对图形而不是考虑数据的分形相似进行了改进，并将结果进行了平滑处理，确保了其预测的精度。 3.提出了一种基于预测的流媒体集群服务器节能策略。该策略根据历史信息，运用加权负载求导预测法预测出系统的未来负载情况，根据各服务器负载状态进行负载转移，休眠多个空闲服务器，实现节能；还可以通过定时唤醒操作，让空闲服务器能够在负载高峰到达前并入系统，满足服务质量要求。因此，该策略可以在保证不对服务能力造成较大影响的前提下，达到节能的目的。 4.搭建了一个具有典型代表性的流媒体集群服务器应用实验平台，实现了本文所提出的加权负载求导预测法和基于负载的服务器节能策略，并验证了其预测精度，还实现了基于预测的负载调度节能策略，证实了该节能控制策略的实际可行性。
[Abstract]:In recent years, with the emergence of a large number of real-time applications, people's requirements for quality of service are also improved. However, when people enjoy the convenience of large-scale cluster server, they ignore the serious problem-energy consumption. With the deepening of the power shortage in China, and the international economic crisis has put forward higher requirements for the management of enterprise operating costs, the improvement of server utilization and energy saving have become the focus of academic and industrial attention. The research in this paper is of great significance for improving the utilization rate of shared network resources and energy-saving control. At present, streaming media applications account for a large share of network applications, so this paper aims at the increasing energy consumption of streaming media servers, and does some work as follows: 1. Under the Linux system, this paper analyzes the hard disk access ability which has not been mentioned in many literatures, improves the method of combining the file system based on / proc and the loading kernel module LKM, realizes the calculation resource, the memory resource, the hard disk resource, and so on. Measurement of network bandwidth resources. The method uses / proc file system to obtain computer information, which is comprehensive, quick and accurate, and the advantages of loading kernel module LKM, such as small resource occupation and high access authority, make it more rapid and accurate. The real-time load of the system is obtained completely. 2. A weighted load prediction method is proposed to predict the network resources of streaming media cluster servers. Based on the daily periodicity of streaming media load, this method makes use of the stability of load curve change rate to predict. In this paper, the existing weighted average calculation methods are improved only for the fractal similarity of the graph rather than the data, and the results are smoothed so as to ensure the accuracy of the prediction. 3. This paper presents a prediction-based energy saving strategy for streaming media cluster servers. According to the historical information, the weighted load prediction method is used to predict the future load of the system, load transfer is carried out according to the load status of each server, and several idle servers are dormant to achieve energy saving. Through regular wake-up operation, the idle server can be integrated into the system before the peak load arrives, to meet the quality of service requirements. Therefore, the strategy can achieve the goal of energy-saving under the premise that it does not have a great impact on service capability. 4. A typical streaming media cluster server application experiment platform is built, and the weighted load derivation prediction method and the load-based server energy saving strategy are implemented in this paper, and the prediction accuracy is verified. Finally, the load scheduling energy saving strategy based on prediction is realized, and the feasibility of the energy saving control strategy is verified.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP393.05

【参考文献】