Fluid limit approximations of bandwidth-sharing networks withrate constraints

Speaker : Masha Frolkova
CWI, Amsterdam
Date: 13/03/2013
Time: 2:00 pm - 3:00 pm
Location: LINCS Meeting Room 40


Bandwidth-sharing networks, as introduced by Massoulie & Roberts (1998), model the dynamic interaction among an evolving population of elastic flows that compete for several links. The main area of application of such models is telecommunications media, e.g. Internet congestion control. With policies based on optimization procedures, bandwidth-sharing networks are of interest both from a Queueing Theory and Operations Research perspective.In this work, we focus on the regime when link capacities and arrival rates are of a large order of magnitude compared to transfer rates of individual flows, which is standard in practice. Bandwidth-sharing networks are rather complicated systems (we operate with measure-valued processes to study them), and under general structural and stochastic assumptions they resist exact analysis. So we resort to fluid limit approximations. Under general assumptions, we derive the fluid limit for the network evolution in the entire time horizon (extending the corresponding result by Reed and Zwart
(2010) for Markovian assumptions). Also, for a wide class of networks, we develop polynomial-time computable fixed-point approximations for their
stationary distributions.