Speaker:     Alex Slivkins, Cornell University
Date:          October 31, 2005
Title:          Oscillations with TCP-like Flow Control in Networks of Queues


We consider a set of flows passing through a network of FIFO servers. The injection rate into each flow is governed by a TCP-like flow control mechanism that increases the injection rate when all the servers on the flow's path are empty and decreases the injection rate when some server is congested. We show that such system can *oscillate*, for (essentially) any given flow control mechanism with instantaneous feedback.

This is in contrast to the previous work on congestion control where congestion on a given server was modeled as a function of the flow injection rates (not the arrival rates at this server), and the system was shown to converge to a steady state that maximizes an overall network utility.

The oscillations in our construction are entirely caused by the queuing dynamics, as the total injection rate into all flows coming through a given server never exceeds this server's capacity.

Joint work with Matthew Andrews (Bell Labs).