Web服务器上比例延迟保证的分散自校正控制方法
发布时间：2018-10-30 13:15
【摘要】：摘要：在服务器端实现比例延迟保证可以显著提高整体网络服务质量(QoS)。Apache是世界使用第一的开源Web服务器软件,可以运行在大多数计算机操作系统上。通过修改源代码,Apache Web服务器可以提供比例延迟区分服务。 Apache Web服务器一般采用单变量模型,其输入为处理不同类别客户端连接工作者线程数量比,输出为不同类别客户端连接延迟比。该单变量模型在处理多优先级延迟保证时需多次建模,可扩展性不佳。针对此情况,提出了一种ApacheWeb服务器多输入多输出模型,并结合分散控制理论和自校正控制理论设计分散自校正控制器。根据分散控制理论,将多变量系统分解为相互关联的单变量子系统,分别对子系统进行辨识得到SISO模型,输入为处理当前类别客户端连接工作者线程数量与工作者线程总量之比,输出为当前类别客户端连接平均延迟与所有类别客户端平均延迟之和之比。为了处理子系统之间的耦合影响,对各子系统应用自校正控制器。该控制器动态调节处理不同优先级请求的工作线程数目,能保证较高优先级请求更快得到处理且维持不同优先级请求的平均延迟比为设定值,模型和控制器参数根据在线辨识结果实时更新。Windows平台下,WinNT MPM是Apache唯一使用的多任务处理模块(MPM)。修改后的WinNT MPM增加了采样机制,连接分类器和控制线程,在Apache内部构建出闭环控制系统。 利用Vmware虚拟机搭建仿真平台,测试结果表明,过载情况下即使并发客户连接数目急剧变化,闭环系统中的服务器仍能维持较好的比例延迟保证。
[Abstract]:Absrtact: the implementation of proportional delay guarantee on the server side can significantly improve the quality of service of the whole network (QoS). Apache is the world's largest open source Web server software can run on most computer operating systems. By modifying the source code, Apache Web server can provide proportional delay differentiation service. The single variable model is generally used in Apache Web servers, whose input is the ratio of the number of worker threads handling different types of clients, and the output is the connection delay ratio of different classes of clients. The single variable model needs to be modeled many times when dealing with the multi-priority delay guarantee, and the scalability is not good. In this paper, a multi-input and multi-output model of ApacheWeb server is proposed, and a decentralized self-tuning controller is designed based on decentralized control theory and self-tuning control theory. According to the decentralized control theory, the multivariable system is decomposed into interrelated single-variable subsystems, and the SISO model is obtained by identifying the subsystems respectively. The input is the ratio of the number of worker threads to the total number of worker threads in the current category client connection, and the output is the sum of the average connection delay of the current class client side and the average delay of all class clients. In order to deal with the coupling effect between subsystems, self-tuning controller is applied to each subsystem. The controller dynamically adjusts the number of worker threads processing different priority requests, which can ensure that higher priority requests can be processed faster and maintain the average delay ratio of different priority requests to set values. The model and controller parameters are updated in real time according to the on-line identification results., WinNT MPM is the only multitask processing module (MPM). Used by Apache under Windows platform. The modified WinNT MPM adds sampling mechanism, connects classifier and control thread, and constructs a closed loop control system in Apache. Using Vmware virtual machine to build a simulation platform, the test results show that even if the number of concurrent customer connections changes sharply, the servers in the closed-loop system can maintain a good proportional delay guarantee.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09;TP273
本文编号：2300112
[Abstract]:Absrtact: the implementation of proportional delay guarantee on the server side can significantly improve the quality of service of the whole network (QoS). Apache is the world's largest open source Web server software can run on most computer operating systems. By modifying the source code, Apache Web server can provide proportional delay differentiation service. The single variable model is generally used in Apache Web servers, whose input is the ratio of the number of worker threads handling different types of clients, and the output is the connection delay ratio of different classes of clients. The single variable model needs to be modeled many times when dealing with the multi-priority delay guarantee, and the scalability is not good. In this paper, a multi-input and multi-output model of ApacheWeb server is proposed, and a decentralized self-tuning controller is designed based on decentralized control theory and self-tuning control theory. According to the decentralized control theory, the multivariable system is decomposed into interrelated single-variable subsystems, and the SISO model is obtained by identifying the subsystems respectively. The input is the ratio of the number of worker threads to the total number of worker threads in the current category client connection, and the output is the sum of the average connection delay of the current class client side and the average delay of all class clients. In order to deal with the coupling effect between subsystems, self-tuning controller is applied to each subsystem. The controller dynamically adjusts the number of worker threads processing different priority requests, which can ensure that higher priority requests can be processed faster and maintain the average delay ratio of different priority requests to set values. The model and controller parameters are updated in real time according to the on-line identification results., WinNT MPM is the only multitask processing module (MPM). Used by Apache under Windows platform. The modified WinNT MPM adds sampling mechanism, connects classifier and control thread, and constructs a closed loop control system in Apache. Using Vmware virtual machine to build a simulation platform, the test results show that even if the number of concurrent customer connections changes sharply, the servers in the closed-loop system can maintain a good proportional delay guarantee.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09;TP273
【参考文献】
相关期刊论文 前3条
1 潘文平;慕德俊;张新家;张慧翔;;相对延迟保证在Apache服务器上的实现[J];计算机工程与应用;2006年29期
2 李捷;刘先省;皇甫伟;;面向Web Server的QoS自适应控制鲁棒性研究[J];计算机工程;2006年11期
3 舒云星;郑卫东;;基于Vmware的虚拟计算机实验系统[J];实验室研究与探索;2006年09期
，本文编号：2300112
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2300112.html
