[e2e] TCP friendly with variable packet size
widmer at informatik.uni-mannheim.de
Tue Jul 1 02:08:45 PDT 2003
> in RFC 3448, the authors indicate that TCP-Friendly rate control
> mechanism should not be used for applications that vary their packet
> size instead of their packet rate in response to congestion.
> Is this true for an application that varies the packet size after the
> estimation of the sending rate and keep this size fixed until the next
> feedback report.
There is a (quite long) technical report about this at
In short: the problem is more with the packet-based measurement of the
loss event rate. If you send smaller packets at a higher rate, you will
underestimate the loss event rate as more and more packet drops will get
aggregated to the same loss event. In the tech report we look at a
number of solutions to remove the bias. However, they are only
applicable if it is ok to trade off packet size for packet rate.
More information about the end2end-interest