NeTS-NR: Preemption and Adaptation for Next Generation Multiservice Networks

该项目的目标是研究在尽力而为网络（如互联网）的背景下，将适应和抢占机制结合使用，作为对拥塞或链路故障发作作出反应的补充工具。特别是，适应被视为对轻度拥塞作出反应的一种手段，而抢占对严重拥塞或链路故障很有用。通过自适应，链路带宽在竞争流之间动态分配，以最大化指定的QoS指标，其中每个流保持其当前路由。然而，通过抢占，流根据分配的优先级级别被选择重新路由。虽然初步的适应和抢占策略都是独立研究的，但本研究项目将这些机制的新组合纳入了一个强大的拥塞控制机制，能够处理不同规模的交通变化。在实践中，自适应本质上是在满足每个流的最小带宽要求后剩余的多余带宽的分布。当拥塞程度太大而无法适应时，抢占是合适的。然而，抢占会对网络稳定性产生潜在的更大影响，并可能导致选择用于重路由的流的服务中断。本研究的目的是在分析、模拟和物理测试平台上研究适应和抢占之间的权衡。结论将是关于各种拥塞制度的最佳拥塞机制的建议。

The objective of this project is to investigate the combined use of adaptation and preemption mechanisms in the context of best-effort networks, such as the Internet, as complementary tools for reacting to the onset of congestion or link failure. In particular, adaptation is seen as a means of reacting to light congestion, while preemption is useful for heavy congestion or link failure. With adaptation, the link bandwidth is dynamically allocated among competing flows so as to maximize specified QoS metrics, where each stream maintains its current route. With preemption, however, flows are selected to be re-routed based on their assigned priority levels. While preliminary adaptation and preemption policies have each been studied independently, this research project includes a novel combination of these mechanisms into a robust congestion control mechanism capable of handling traffic changes of varying magnitude. In practice, adaptation is essentially a distribution of the excess bandwidth left over after each stream's minimum bandwidth requirements have been satisfied. Preemption is appropriate when the congestion level is too great for adaptation to be viable. Preemption, however, carries a potentially greater impact on network stability and can cause service disruption for the flows selected for rerouting. The goal of this research is to study the tradeoffs between adaptation and preemption both analytically, in simulation, and on a physical testbed. The conclusion will be a recommendation regarding the optimal congestion mechanism for various congestion regimes.