Predictive routing to enhance QoS for stream-based flows sharing excess bandwidth

We propose a new routing algorithm based on online estimation of the link load dynamics and prior information on flow holding times. The motivation for this proposal lies in supporting traffic flows such as VBR or associated with rate adaptive applications. Such traffic requires a minimal guaranteed bandwidth, but can see improved performance when sharing excess bandwidth not used to meet guarantees. The key idea underlying our approach is to route traffic flows so that they see minimal expected flow-perceived loads during their sojourn in the network. To this end we establish a routing framework where links estimate and advertise the parameters associated with their load dynamics in addition to their current load. New flows are routed based on this information and prior knowledge of their holding times so as to minimize the expected flow-perceived load. Simulations of this routing scheme in a (weighted) max–min bandwidth sharing framework show gains of 20–50% in the average flow bandwidth share over baseline routing schemes. 2002 Elsevier Science B.V. All rights reserved.