CAREER: Integrated Traffic Management Based on Accurate and Practical Packet Fair Queueing Algorithms

多种网络服务包括保证实时服务、预测实时服务和尽力而为服务。虽然已经有大量研究为每个单独的服务设计流量管理算法，但尚不清楚这些算法如何集成到一个网络中，以及它们将如何相互作用。当网络还需要支持链路共享时，这个问题变得更加复杂，链路共享允许在需要不同网络服务但属于同一管理类的应用程序之间共享资源。交换节点上的服务规则决定了为分组提供服务的顺序，在控制不同业务流、不同服务类别和不同链路共享类别之间的交互方面起着关键作用。假设的流体公平排队(FFQ)规则可以根据预先指定的服务份额同时服务于多个业务流，为集成针对不同服务和链路共享的流量管理算法提供了理想的框架。由于FFQ使用的是无法在现实世界中实现的理想化流体模型，因此需要使用分组公平排队(PFQ)算法来近似其行为。以前提出的PFQ算法，包括最流行的加权公平排队算法，都在近似FFQ时引入了很大的误差，这严重影响了流量管理和链路共享算法。目前还不清楚是否存在准确的PFQ，如果存在，它们是否足够简单，可以高速实现。这项研究的目标是证明可以设计准确和实用的PFQ算法，并且它们为框架提供基础，在该框架下可以以集成的方式支持针对不同服务和链路共享的流量管理算法。所涉及的方面包括(A)PFQ算法的复杂性和准确性之间的基本折衷；(B)将不同的业务管理算法和链路共享集成到一个框架中的问题以及在PFQ算法的准确性和链路共享和业务管理算法的效率之间的基本折衷；(D)用于在不仅在速率成比例而且在一般资源分配的情况下使用PFQ算法来提供端到端性能保证的接纳控制算法；e)更灵活的实时服务，例如基于PFQ算法的重新协商服务和速率自适应服务。这项研究通过分析、模拟和实施相结合的方式进行。这项工作的结果将对互联网和ATM环境下的综合流量管理算法的设计产生重大影响。

Multiple network services include guaranteed real-time service, predicted real-time service, and best-effort service. While there has been extensive research on designing traffic management algorithms for each individual service, it is unclear how these algorithms can be integrated into one network and how they will interact with each other. This problem becomes more complex when the network also needs to support link-sharing, which allows resource sharing among applications that require different network services but belong to the same administrative class. The service disciplines at the switching nodes, which determine the order in which packets are serviced, play a critical role in controlling the interactions among different traffic streams, different service classes, and different link sharing classes. The hypothetical Fluid Fair Queueing (FFQ) discipline, which can serve multiple traffic streams simultaneously according to pre-specified service shares, provides an ideal framework to integrate traffic management algorithms for different services and link sharing. Since FFQ uses an idealized fluid model that cannot be implemented in the real-world, Packet Fair Queueing (PFQ) algorithms are needed to approximate its behavior. Previously proposed PFQ algorithms including the most popular Weighted Fair Queueing all introduce substantial inaccuracy in approximating FFQ, which significantly affects traffic management and link sharing algorithms. It is unclear whether accurate PFQ's exist, and if they do, whether they can be simple enough to be implementable at high speed. The goal of this research is to demonstrate that accurate and practical PFQ algorithms can be designed and that they provide the foundation for a framework under which traffic management algorithms for different services and link sharing can be supported in an integrated fashion. Aspects treated include (a) the fundamental tradeoff between complexity and accuracy of PFQ algorithms; (b) issues in integrating different traffic management algorithms and link-sharing into one framework and fundamental tradeoffs between accuracy of PFQ algorithms and the efficiency of link-sharing and traffic management algorithms; (d) admission control algorithms for providing end-to-end performance guarantees using PFQ algorithms under not only rate-proportional but also general resource assignments; e) more flexible real-time services such as renegotiated service and rate-adaptive service based on PFQ algorithms. The study is conducted via a combination of analysis, simulation, and implementation. The results of this work will have a significant impact on the design of integrated traffic management algorithms in both the Internet and ATM environments.