Admission Control in High Speed Multimedia Networks

近年来，技术的进步和带宽的可用性带来了一个新的多媒体和实时应用网络，如互联网电话、互动电视、视频点播和视频会议。这些应用程序在从教育、医疗保健到娱乐的广泛活动中具有重要用途。它们的全面部署显然将对我们的生活质量和整个社会产生巨大影响。然而，这样的应用程序对带宽要求很高，对拥塞非常敏感，并且需要实时可靠的通信。尽管已经存在适当的协议，例如异步传输模式（ATM）协议和Internet的TCP/IP协议的增强版本（带有RSVP），但服务质量（QoS）供应仍然是需要在可靠地支持实时应用程序之前解决的基本开放问题之一。如果不能选择拥塞，那么限制网络中允许的用户数量将成为保证QoS的非常有用的工具。本研究的中心目标是设计这样一个接纳控制程序，并在实践中测试其性能。为了提供QoS保证，网络应该防止由于缓冲区溢出和过度延迟而导致数据包丢失的拥塞。基于最坏情况的准入控制算法（分配给呼叫的带宽等于其峰值速率）会导致网络资源的严重利用不足，因为流量通常是突发的，即峰值速率不会持续很长时间。为了更有效地利用资源，建议的方法将允许出现一些拥塞现象，这些现象发生的频率不高，足以提供统计QoS保证。更具体地说，希望在损失概率和过度延迟概率特别小（例如，在10^{-6}的数量级上）的情况下运行网络。实时应用程序可以容忍如此小频率的拥塞现象。这项研究将借鉴应用概率、优化和最优控制理论中发展起来的方法，并在必要时扩展PI最近发展起来的许多思想。计算非平凡交通模型的拥塞概率尤其困难。因此，建议的工作将采用大偏差理论（应用概率的一个分支）作为主要分析工具的渐近结果。被认为是成功入学控制的先决条件的问题将首先得到解决。其中提出的研究将涉及：-交通建模。目的是提出能够捕获与接纳控制相关的流量统计特性的模型。性能分析。目标是渐近估计多类多跳网络环境中的拥塞概率。建议工作的最终产品将是一个允许控制算法，它为被允许的调用提供统计QoS保证，并具有以下特征：-支持多个服务类。应用程序将根据它们产生的流量的统计特征和它们的QoS要求被捆绑在服务类别中。-充分利用可用带宽。控制器将具有足够的灵活性来研究服务类之间不同的带宽分配策略，并且只有在没有可行的分配保证所有连接呼叫的QoS时才拒绝允许。-端到端QoS保证。实时实现。关于这个项目的更多信息可以从http://ionia.bu.edu/网站上获得

In recent years, technological advances and bandwidth availability have brought about a web of new multimedia and real-time applications, such as internet telephony, interactive TV, video-on-demand, and videoconferencing. These applications have important uses in a wide range of activities from education and health care to entertainment. Their full deployment will clearly have a huge impact on the quality of our lives and the society as a whole. Such applications, however, are very bandwidth demanding, very sensitive to congestion, and require real-time and reliable communication. Although appropriate protocols, such as the Asynchronous Transfer Mode (ATM) protocol and an enhanced version of Internet's TCP/IP protocol (with RSVP), already exist, Quality of Service (QoS) provisioning remains one of the fundamental open problems that need to be addressed before real-time applications can be reliably supported. If congestion is not an option, restricting the number of users that are allowed in the network becomes an extremely useful tool in guaranteeing QoS. The central goal of this research is to devise such an admission control procedure and test its performance in practice. To provide QoS guarantees the network should prevent congestion which causes packet losses due to buffer overflows and excessive delays. A worst case based admission control algorithm (which allocates to calls bandwidth equal to their peak rate) leads to significant underutilization of the network resources, since traffic is typically bursty, i.e., the peak rate is not sustained for long. To achieve more efficient use of the resources the proposed approach will permit some congestion phenomena which will be occurring infrequently enough to allow for the provision of statistical QoS guarantees. More specifically, it is desired to operate the network in a regime where loss probabilities and probabilities of excessive delays are particularly small (e.g., on the order of 10^{-6}). Real-time app lications can tolerate such small frequencies of congestion phenomena. This research will draw upon methods developed in applied probability, optimization, and optimal control theory, and will use, extending whenever necessary, many of the ideas developed recently by the PI. Calculating congestion probabilities for non-trivial traffic models is particularly hard. The proposed work will therefore resort to asymptotic results using large deviations theory (a subfield of applied probability) as the main analytical tool. Issues that are considered as prerequisites for successful admission control will be first resolved. Among them the proposed research will address: -Traffic Modeling. The aim is to propose models that capture the statistical properties of the traffic which are relevant to admission control. -Performance analysis. The goal is to asymptotically estimate congestion probabilities in a multiclass and multihop network environment. The final product of the proposed work will be an admission control algorithm that provides statistical QoS guarantees to admitted calls and possesses the following characteristics: -Support for multiple service classes. Applications will be bundled in service classes based on the statistical character of the traffic they generate and on their QoS requirements. -Full utilization of available bandwidth. The controller will be flexible enough to investigate different bandwidth allocation strategies among service classes and deny admission only when there is no feasible allocation that guarantees QoS to all connected calls. -End-to-End QoS guarantees. Real-time implementation. More information about this project can be obtained from the Web at http://ionia.bu.edu/