Flash Crowd Effect in RCP

The Rate Control Protocol (RCP) [1] is an explicit congestion control mechanism that, amongst other characte ristics, reduces the average flow completion time (AFCT) metric by one order of magnitude when compared to TCP NewReno. RCP reduces the AFCT by allowing new flows to instantly use the same rate as existing flows in the network. This results in link utilization temporarily exceeding available capacity when flows enter a network, inducing queue build-up. As such, RCP is particularly vulnerable to flash crowds , whereby a system witnesses a significant increase in the number of flows over a short period of time. In this paper we analyze RCP’s response to varying rates of increase in the number of flows. We conclud e that, for a given arrival growth rate, RCP is able to stabilize queue length as long as this rate does not exceed well defined limits. We quantify the queue length required to stabilize the system response and the limit arrival growth rate using a model of RCP that incorporates the effect of new flow arrivals. Finally, we validate our analysis through ns-2 simulations.