Intra-Flow Fairness in Work-Conserving Flow Aggregation

J.A. Cobb and Z. Xu (USA)

Keywords

Quality of Service, Real-Time Scheduling, Flow Aggrega tion.

Abstract

Currently, the Internet provides mainly best effort services. As multimedia applications emerge, a demand for qual ity of service guarantees follows. Packet-scheduling pro tocols, such as Virtual Clock and Weighted Fair Queuing, were designed to fulfill this demand by reserving network resources for each individual packet stream (also known as flows). This leads to a scalability problem in the core of the network, because routers are expected to maintain state information for each individual flow. One method to mit igate this problem is to merge together into a single flow those flows following a common path through the core. This drastically reduces the number of flows managed by a router. Initial flow aggregation methods were non-work con serving. This prevents flows from taking advantage of un used bandwidth, that is, from exceeding their reserved rate when the network is lightly loaded. Recently, this was remedied with the introduction of work-conserving flow aggregation methods. However, these methods are unfair, i.e., a flow that takes advantage of unused bandwidth may temporarily be denied service from the network. In this paper, we propose a flow aggregation method that is fair, and thus, flows are not punished if they exceed their reserved rate. We show that this new method still con serves the end-to-end delay guarantee of work-conserving flow aggregation. Moreover, because of its fairness, it also provides an end-to-end throughput guarantee.

Important Links:



Go Back