【作者】 王俊华；

【导师】 林政宽；

【作者基本信息】 苏州大学 ， 计算机科学与技术（专业学位）， 2015， 硕士

【摘要】 随着互联网的飞速发展,网络承载的流量和内容也越来越庞大,由此造成的网络拥塞和用户体验下降也越来越成为人们关注的焦点。由于如今网络流量的使用量过大,以及各种P2P,VOIP等这类特征特殊且占据网络大量带宽流量的应用类型的产生,使得提高网络流量传输质量成为必然要解决的问题。传统情况下,人们往往通过增加网络处理器等来提高性能,然而硬件具有可编程性复杂与无法承担复杂任务等缺点,无法适应日趋灵活的网络流量需求。要提高网络传输质量,一方面需要尽可能利用现有的硬件体系提高性能。在这点上,本文采取了一种多核并行式计算机结构。自多核处理器被发明以来,其发展就广受瞩目,随着一个个难点被解决,其体现出来的性能是传统网络处理器专用硬件不能比拟的。另一方面,对流量进行一定的管理也能提高网络传输质量,本文所采用的方案是使一些无弹性与时延要求较高的流量不会被另一些弹性强与重要性较低的流量所冲突,这个方法需要建立一个良好的QoS框架。本文着重讨论了如何在满足大流量带宽控制需求的高性能的多路/核的CPU硬件体系上实现并行、无锁的QoS程序设计和算法实现。本文首先介绍了QoS的概念与技术要点,然后对QoS的框架作了一定的说明,并介绍了用到的识别与流量整形的算法,流量识别分类的嵌套结构。接着对于如今的多核并行式系统进行了说明。由于并行式系统的一大难点就是解决线程与线程之间的冲突问题,本文针对程序设计中的难点提出了解决方案,利用无锁编程以及用原子操作替代锁操作的方式来解决串行式问题。最后重点针对如何将解决方案实际应用到流控框架中并实际对流量进行监管的整个数据流处理过程进行了详细的解说,该控制算法最后被证明在真实环境中是稳定有效的。更多还原

【Abstract】 Along with the rapid development of Internet, the network payloads are becoming heavier, many researchers focused on the cause of network congestion and user experience.Nowadays, the large usage of network traffic, such as P2P, VOIP, and other kind of streams like this, has occupied much bandwidth of network traffic, making the quality of transmission network traffic a problem that urged to be solved. Traditionally, people tend to improve performance through the network processor. However, programming for the network hardware is a complicated problem, and unable to handle the tasks that too complex, which made it hard to meet the increasingly flexible network traffic demands.If we want to improve the quality of network transmission as much as possible, on the one hand, we need to make good use of existing hardware system to improve performance. At this point, this paper adopted a multi-core parallel computer architecture. Since multi-core processors was invented, and with many difficulty is solved, its performance has already achieved a level that network processor can’t compared with. On the other hand, we also need to control the management of the network flow, so that some inelastic, high demand on latency flow will not be conflicted and jammed by another elasticity, less important traffic, for this, we design an QoS framework for this.This paper focused on how to meet the traffic control requirements on high-performance multiplex/core CPU hardware system and achieve the parallel, lock-free QoS program design and algorithm implementation. First, this article introduced the concept of QoS and technology that related, then introduced the framework of the QoS, and what algorithms QoS used for identification and traffic shaping, and the nested structure of flow recognition classification. After that, we introduced the parallel system and its problems, which is the conflict between two threads. This paper puts forward the solutions, that is using the lock-free programming in some place and replace the lock with atomic operation in other place to solve the problem. Finally the paper explained how to run this design on real QoS framework and the whole process of data stream being dealed with, the control algorithm in the real environment is proved to be steady and effective.更多还原