【作者】 陈晓峰；

【导师】 平玲娣； 潘雪增；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2006， 博士

【摘要】 互联网能迅速发展到现在的规模,很大程度上要归功于其TCP/IP协议簇“尽力而为”(Best-Effort)的设计思想,大大降低了网络核心的复杂度和负担,使其能够迅速扩展,同时这一做法也使得网络缺乏对服务质量(QoS)的保证。随着网络规模的不断增大,各种各样的网络服务争相涌现,互联网已逐步向传输数据、语音、图像等多媒体信息的综合传输网演化,与此同时,传统电信业务的承载平台也正在向基于IP技术的分组交换系统转化,IP网络缺乏QoS保证的缺点正日益突出。在这种背景下,IP QoS已经成为未来IP网络发展的关键技术。 本文回顾了TCP/IP网络的原理,给出了IP网络QoS的定义以及实现思路和手段,在相关研究工作和进展的基础上,做出了如下创新工作: 1、针对下一代互联网研究计划Internet2提出的Scavenger Service(SS)的特点,提出了一种实现SS的对数自适应队列调度算法LAWRRQ。该算法用滑动时间窗口算法统计活跃流数量,并据此以对数的规律自适应地调整带宽在SS流和Best-Effort(BE)流之间的分配。算法利用对数函数的特性,在流数量统计精度和开销之间找到了平衡;利用流数量和总轮转片数对照表的方式克服了计算复杂性;针对SS数据包在缓冲区中堆积的现象加入了对BE队列的保护算法。实验结果表明,该算法在很好地保护BE流的同时,为SS流提供了更可靠的最小带宽保证,同Internet2推荐的现有队列调度算法相比,具有更好的调度性能和鲁棒性。 2、通过对SS聚合流内部公平性的研究,指出了SS不适用于非适应流的问题。随后研究了用于平衡带宽的FRED队列管理算法在SS下的性能,发现了对适应流存在较多误判的问题,并对其做出了改进,提出了MFRED算法,在算法中增加了一种误判纠正机制。实验结果表明,MFRED能有效平衡带宽在SS流之间的分配,明显减少了对适应流的误判,使得SS的适用范围从TCP等适应流扩展到UDP等非适应流。通过对MFRED中nactive参数值和实际SS流数量的关系的研究,提出了一种用nactive值近似估计活跃SS流数量并据此调节最低带宽保证的方法,并在实验中得到验证。 3、提出了一种确保服务中实现聚合流带宽按协议承诺信息速率比例公平分配的IEAM标记算法,将TCP中的AIMD控制思想引入到聚合流标记速率的调整上面。在该算法中,核心路由器通过检测IN包平均队列长度,标记反方向转发的数据包,进行直接拥塞信息反馈;边界入口路由器根据拥塞反馈信息采用一种AIMD的方式进行标记速率的调整。该算法和已有方案相比,具有简捷、自适应性和可扩展的特点。模拟实验表明,IEAM标记算法在不同网络条件下,更多还原

【Abstract】 The reason why Internet can expand to such an enormous scale today mostly contributes to the Best-Effort design principle in TCP/IP protocol stack. This principle greatly brings down the complexity and the burden of the network core and makes it able to scale up quickly. As the fast expanding of network scale, various new network services emerge. Internet is now changing into an integrated network that transmits both data and multimedia flows. Also, the carrying foundation of traditional telecom applications is shifting into packet switching network based on IP technology. The shortcoming that original IP network lacks QoS (Quality of Service) warranty becomes increasingly critical. How to achieve QoS in IP network has become a key issue in next generation network development.This dissertation first reviews the elements of TCP/IP network and gives the definition of IP QoS together with the way to its realization. Then, on the basis of relative research works and progresses, the dissertation presents following innovative works:1. To deal with the particularity of Scavenger Service (SS) from Internet2 project, a logarithmically adaptive queue-scheduling algorithm named LAWRRQ is presented. In the algorithm, the number of active Scavenger Service flows is calculated by a time sliding window algorithm, and the bandwidth allocated for Best-Effort (BE) flows and SS flows is accordingly tuned in a logarithmical way. The tradeoff between flow number counting accuracy and cost is made by the characteristic of the logarithmic function. A looking-up table composed of flow number and total round robbing slice number resolves the computing complexity. A protection for BE queue in buffer management is also introduced. Simulation results show that compared with current algorithms recommended by Internet2, the algorithm provides more reliable bottom bandwidth guarantee for Scavenger Service flows when protecting BE flows well, and has better performance and robustness.2. The fairness between Scavenger Service micro-flows is studied. The problem that Scavenger Service is not applicable for non-adaptive flows is pointed out. Then, the study on performance of FRED queue management algorithm dealing with Scavenger Service shows the problem of many misidentifications with adaptive flows. Based on this, A MFRED algorithm is presented, in which a mechanism designed for misidentification correcting is added. Simulation results show that MFRED identifies non-adaptive flows and balances the bandwidth allocation effectively while misidentifications being greatly reduced. By means of study on therelationship between parameter nactive and actual active SS flow number in MFRED, a method of estimating SS flow number by nactive to tune the guarantee bottom bandwidth of Scavenger Service is presented and identified.3. An IE AM marking algorithm for proportional bandwidth distribution of Assured Service in DiffServ is presented. In the algorithm, the AIMD control thinking from TCP is introduced into the aggregate flow marking rate tuning. The core routers judge the congestion level according to the average length of IN queue and send out feedback information by marking the packets being forwarded to inverse direction at the moment of congestion. Based on the feedback information, the edge routers adjust their target rates in an AIMD way without communications with other edge routers. The presented algorithm is simple, adaptive and scalable. Simulations shows that the proposed algorithm realizes proportional bandwidth allocation according to committed information rate without communications between different edge routes under different network conditions.4^ Current research works on wireless mobile IP QoS are Summarized and the different schemes are compared with each other. These works are divided into four aspects: management of resource reservation, re-establishment of reservation path, design of mobile agent and integration of RSVP and mobility management. Based on some mechanisms mentioned in summarization, a QoS guarantee scheme in Fast Handovers for Mobile IPv6 architecture is presented. Finally, some basic simulations are made.更多还原