Resource management in computer networks

计算机、通信和网络技术的进步使得许多网络应用成为可能。一类重要的应用程序需要及时交付实时数据。传统的基于IP的网络提供尽力而为的服务;当流量很小时，它们的延迟性能很好。然而，当流量增加时，例如当数据路径沿着发生拥塞时，它们不能很好地扩展，因此为了支持实时应用，需要网络服务质量（QoS）策略。示例策略是信道调度、准入控制和定价。在过去的二十年里，两种方法来提供网络QoS已经开发：基于流和基于截止日期。在这项研究中，我们将设计和评估一个统一的方法来提供网络QoS的解决方案和调查结果的基础上，从基于流和基于截止日期的方法。我们的目标是开发具有成本效益的端到端QoS解决方案，支持各种实时应用。QoS提供中的关键参与者是网络（或因特网）服务提供商（ISP）。在本研究中，我们也针对ISP服务品质策略的设计。与端到端的QoS研究不同，本文的重点是ISP网络中的资源管理。这些研究的结果将有助于下一代计算机网络的QoS基础设施的设计。一个设计良好、规模合理的网络应该能够容纳大部分流量。数学模型可用于网络设计和尺寸确定。已经为类似的网络开发了良好的数学模型，并且可以应用于基于IP的网络，只要呈现某些业务特性。这样的业务特性可以通过影响业务特性的那些实体的适当设计来获得。示例实体包括较高层协议、网络资源管理和业务复用。我们研究这些实体对流量的影响;我们的长期目标是设计这些实体，并为端到端的数据通信提出合理的性能模型。这项研究的结果将提供宝贵的洞察下一代计算机通信网络的性能建模。

Advances in computer, communication, and networking technologies have enabled many network applications. An important category of applications requires timely delivery of real-time data. Traditional IP-based networks provide a best-effort service; their delay performance is good when the traffic is light. However, they do not scale well when the level of traffic increases; for example, when congestion occurs along the data path. Thus to support real-time applications, network quality-of-service (QoS) strategies are needed. Example strategies are channel scheduling, admission control, and pricing. Over the past two decades, two approaches to network QoS provision have been developed: flow-based and deadline-based. In this research, we will design and evaluate a unified approach to network QoS provision, based on solutions and findings obtained from both flow-based and deadline-based approaches. Our goal is to develop cost-effective end-to-end QoS solutions that support a wide variety of real-time applications. A key player in QoS provision is the network (or Internet) service providers (ISPs). In this research, we also target the design of ISP QoS strategies. Differ from the end-to-end QoS study, the emphasis here is the resource management within an ISP network. Results from these studies would aid in the design of QoS infrastructure for next generation computer networks. A well-designed and well-dimensioned network should accommodate most of its traffic. Mathematical models may be used in network design and dimensioning. Good mathematical models have been developed for similar networks and may be applied to IP-based networks, provided that certain traffic characteristics are presented. Such traffic characteristics may be obtained by the proper design of those entities that affect traffic characteristics. Example entities include higher-layer protocols, network resource management, and traffic multiplexing. We study the effect of such entities on traffic; our long-term objective is to design such entities and to come up with sound performance models for end-to-end data communication. Results from this research will provide invaluable insight into the performance modeling of next generation computer communication networks.