Extend of my previous work. Consider stochastically TCP-friendliness that defined as:

where $\pi$ denotes the control for UDP traffic and $X$ are the rates of a set of sessions; $U$ is a set of utility function and the comparison $\le$ is component-wise to the vector $X$.

If $U$ is defined as the set of all increasing functions ($U=st$), then the stochastic ordering $X^\textrm{TCP} \le_{st} X^\pi$ is the usual stochastic order, which means

The protocol in this paper, TCP-Friendly CBR-Like Rate Control or TFCBR, is not to apply admission control but to smooth the rate of sending to a longer time scale, say, minutes. There are two parameters in this scheme, namely, $\alpha$ the fraction of TCP rate to send by the TFCBR; and $\beta$ the time constant. The protocol checks TCP rate for previous $\beta$ seconds using equation and adapt to the $\alpha$ of that rate. Thus it is like a moving average and it claims that, by appropriately adjusting the parameters, it can be stochastically TCP-friendly.

The paper also points out, $\alpha$ cannot be greater than 1 or otherwise it is not TCP-friendly at all.

Bibliographic data

   title = "TCP-Friendly CBR-Like Rate Control",
   author = "Feng and Xu",
   booktitle = "Proc ICNP",
   year = "2008",