The Internet traditionally provides best effort service to all applications. While elastic applications are satisfied by this service, inelastic applications such as interactive audio and video suffer from end-to-end delay guarantees. Although Guaranteed Rate schedulers were developed to provide such guarantees, their scalability has been a concern because they maintain per-flow state. In an effort to reduce per-flow state, two methods have been proposed: stateless core networks and flow aggregation. Stateless core networks require no per-flow state at the routers, while flow aggregation maintains state for a small number of aggregate flows. Although flow aggregation maintains more state, it provides a lower end-to-end delay bound than stateless core networks. The original proposals of these two techniques did not provide guaranteed throughput, that is, flows could be temporarily denied service if they exceeded their reserved rates at earlier times. Recently, guaranteed throughput has been incorporated into the stateless core model through the reuse of deadlines. This is similar to the deadline reuse found in earlier stateful protocols that provide guaranteed throughput. In this paper, we propose adding deadline reuse to flow aggregation networks. In this way, guaranteed throughput can be achieved while maintaining a lower end-to-end delay bound. In addition, we revise the deadline reuse method for stateless core networks.
展开▼
