AW: AW: [e2e] Queue size of routers

[Saverio Mascolo]

I would like to add one point to the discussion: the point is that the
optimal queue size depends also on the TCP congestion control algorithm,
that is, C*T bytes of buffers are required by senders using classic Reno/New
Reno algorithms but not by senders implementing the additive increase
adaptive decrease paradigm such as   Westwood TCP (see below)

> The previous result is related to the "sawtooth" variations of TCP's
> congestion window, and it can be roughly explained as follows:
> after a (single) packet loss cwnd will be decreased by
> a factor of 2. To keep the link saturated after the loss,
> we need to maintain a send window of at least C*T, which means
> that before the loss cwnd should be at least 2*C*T. This will be the
> case if we have C*T bytes in the buffer of the bottleneck link,
> and C*T more bytes in the "wire" (please send me a note if
> you would like to receive a (still incomplete..) draft that
> discusses this problem in more detail).

what Constantine says is based on the assumption that the sender implements
a classic by half window reduction after packet loss. However if you assume
a TCP Westwood sender, which adaptively reduces the cwnd to the value C*Tmin
, then the C*T bytes of buffers are no more required: in fact, the window is
not "blindly" reduced by half but it is exactly reduced of the requested
amount.

A simulation  considering  a single connection with T=RTT=250ms over a 10
Mbps bottleenck link has shown that the throughput/goodput of (New)Reno
increases with the queue size and saturates when the queue size approaches
the C*T size.
On the other hand TCP Westwood is able to provide full link utilization also
with queue size largely smaller than C*T. In particular, with a queue size
of 0.1*C*T, Westwood reaches practically full link utilization (9.5 Mbps)
whereas Reno reaches 7.5Mbps.

cheers

Saverio Mascolo