Work-Conserving Fair-Aggregation with Rate-Independent Delay

Abstract

Flow aggregation has been proposed as a technique to improve the scalability of QoS scheduling in the core of the Internet, by reducing the amount of per-flow state necessary at a router. This is accomplished by aggregating into a single flow multiple flows that share part of the same path to the destination. In its simplest form, flow aggregation is inherently non-work- conserving, that is, it may temporarily leave the output channel idle even though there are still packets to be forwarded. Work-conserving variations of flow aggregation are more desirable, because they allow flows to temporarily exceed their reserved rate and take advantage of bandwidth unused by other flows. This, however, requires a per-hop delay proportional to the rate reserved for the aggregate flow. Thus, if a lower delay is desired, the aggregate flow must reserve from the network a data rate that is greater than its true rate. In this paper, we explore the conditions under which work- conserving flow aggregation can be performed with rate- independent delay. That is, the aggregate flow is able to choose a per-hop delay that is independent of the data rate it reserved from the network. The price to be paid for this flexibility, however, is a restriction in the choice of packet size and data rates of its constituent flows.