RE: TCP (and SCTP) sucks on high speed networks

["Bernard Aboba" <aboba@internaut.com>]

>Consider a 10Gbs link to a destination half way around the world.  A packet
>drop due to link errors (not congestion or infrastructure products) can be
>expected about every 20 seconds.  However, with a RTT of 100ms (not even
>across the continent), if a TCP connection is operating at 10Gbs, the
packet
>drop (due to link error) will drop the rate to 5Gbs.  It will take 4
*MINUTES*
>for TCP to ramp back up to 10Gbps.

I dispute the basic premise here. At the speeds of 10 Gbps, we are talking
about optical transport which is immune from electrical noise. With such
a large window size, my assertion is that the vast majority of packet drops
would in fact be due to congestion, not "link errors".

The thing to be concerned with here is not the behavior of a single
TCP/SCTP connection but rather the behavior of an ensemble of connections.
In general, Storage Area Networks are used to connect many hosts to
a storage controller. So in practice you will have hosts competing
with each other for bandwidth and you need a transport with the
ability to adjust to available bandwidth.

As a result, TCP/SCTP behavior is exactly what you want.