Flow level dynamics of congestion control for file transfers

Laurent Massoulie
Microsoft Research

In this talk, we consider the dynamics of file transfers sharing network resources via congestion control. We propose a stochastic model for these dynamics motivated by a time scale separation assumption, namely that congestion control achieves target rate allocations in a time short compared to that over which the population of ongoing file transfers changes.

We establish that the corresponding stochastic process has the largest possible stability region provided the target rate allocations are fair. We show that this no longer holds with unfair allocations, e.g. when flows of different kinds have prioritised access to bandwidth.


References:

T. Bonald and L. Massoulie Impact
of fairness on Internet performance
ACM Sigmetrics 2001.


Back to Long Programs