Adaptive and Scalable Quality of Service Provisioning in Multirate Multicast Networks

多播在实时应用中变得越来越重要，因为它是其中一些应用的固有传递方式，例如电话会议、音频和视频广播等。然而，Internet并不为单播或多播数据提供端到端带宽或延迟保证。因此，针对实时语音和视频传输的多尺度编码方案被开发出来；这些方案允许通过改变信息流的级别数量和相应的比特率来适应可用带宽和接收器的能力。因此，带宽和接收器对多级信息流的适应，特别是在多播传输中，对于这些实时应用的成功大规模部署是一个至关重要的问题。我们建议在基于多播的实时应用中研究和开发带宽共享和适应的方法。提出了一套公平性准则，以此来量化和比较不同方案的性能。提出了一种根据不同标准寻找最优或次优带宽分配的方法。在评估不同算法时，除了最优性外，网络规模的可扩展性、所需控制信息对分布式实现的适应性和收敛速度也被作为主要标准。我们提出了一种算法设计方法，允许在性能目标的最优性和其他属性（即可扩展性等）的实用性之间进行权衡。本文还提出了一种在流量和/或网络条件未知的情况下进行多播网络控制的方法。它包括明确的基于背压的拥塞通知反馈和动态流量调度。最后，我们提出了一种以最大化总网络效用为目标的拥塞控制策略设计方法。还提出了两种方法（基于公平和总效用最大化）之间的相对比较，以便能够在每种情况下选择最合适的方法。

Multicasting has become increasingly important for real-time applications since it is the inherent mode ofdelivery in several of them, e.g. teleconferencing, audio and video broadcasting, etc. The Internet, however,does not provide end-to-end bandwidth or delay guarantees for unicast or for multicast data. As a result, anumber of multiple scale encoding schemes for real-time voice and video transmission have been developed;these schemes allow adaptation both to the available bandwidth and to the receiver capabilities by varyingthe number of levels of the information stream and therefore the corresponding bit rates. Thus, bandwidthand receiver adaptation of multi-level information streams, in multicast delivery in particular, is an issueof prime importance for the successful, large-scale deployment of those real-time applications. We proposeto investigate and develop methods for bandwidth sharing and adaptation in the context of multicast-based real-time applications. A set of fairness criteria based on which the performance of different schemescan be quantified and compared is introduced. A methodology for finding bandwidth allocations that areoptimal or suboptimal with respect to the different criteria is proposed. Scalability with the network size,amenability to distributed implementation in terms of required control information and speed of convergenceare adopted as prime criteria, in addition to optimality, in the evaluation of the different algorithms. Wepropose an approach for the design of the algorithms that allows the trade-off between optimality withrespect to the performance objective and practicality with respect to the other attributes (i.e. scalabilityetc.). A methodology for network control in multicasting in the case of unknown traffic and/or networkconditions is proposed as well. It involves explicit back-pressure based congestion notification feedback anddynamic per ow scheduling. Finally we propose a method for the design of congestion control policieswith the objective of maximizing an aggregate network utility. A relative comparison between the twoapproaches (fairness based and aggregate utility maximization) is also proposed in order to be able to selectthe most appropriate one in each case.