[CST-2] D CommII and TinC
Phebe Mann
pm258@hermes.cam.ac.uk
Mon, 4 Jun 2001 21:18:32 +0100 (BST)
On Mon, 4 Jun 2001, Shu Yan Chan wrote:
> I got some little questions which I do not managed to clear it up over the
> year, I hope it is not too late to ask:
>
> Digital Communication II:
>
> P.44 and 46 of Steven Hand notes, the graphs of cwin against time. Why
> after the first time out and the cwin drops, cwin does not seem to be under
> niether linear nor expoential increase? In P.44 Steve Hand said it is in
> congestion avoidance after 2ms, but why we still got packet loss???
>
TCP Reno's congestion detection and control mechanism is reactive rather
than proactive, it needs to create losses to find the available bandwidth
of the connection. The mechanism to detect the available bandwidth is to
continually increase its window size using up buffers along the
connection's path until it congests the network and segments are lost. It
then detects these losses and decreases it window size.
But for TCP Vegas
The expected throughput is given by
Expected = WondowSize / BaseRTT
It calculates the currect Actual sending rate, by recording teh sending
time for a ditingushed segment, recording how many bytes are transmitted
between the time that segment is sent and its ack is received, compute the
RTT for the distinguishe segment, whne the ack arrives, divide the number
of bytes transmitted by teh sample RTT. (calculate one per RTT)
Compare Actual to Expected, adjusts teh window accordingly.
So TCP Vegas endeavours to prevent packet loss.
Phebe