[e2e] High Packet Loss and TCP

[Jon Crowcroft]

we saw lots of horrible behaviour later when the link was stressed but the period when we
had a full line rate monitor was rather uninteresting tcp-wise - ian's right about dns
(though we didn't write it up much - its prob. a forerunner of later dns-is-a-disaster
work) - 

the later work (which motivated congestion pricing and other tcp hackery (multcp etc) as
well as the UK natioanl cache work, was more done by UKERNA and peter linington i belive.

of course later still , the price of long haul capacity in bulk (at least)
 fell thru the floor so the problem went away...

In missive <16051.29065.96996.897772 at harp.crn.cogs.susx.ac.uk>, Ian Wakeman typed:

 >>>>>>> "Craig" == Craig Partridge <craig at aland.bbn.com> writes:
 >>
 >>    Craig> Didn't Jon Crowcroft capture some traces from the
 >>    Craig> trans-Atlantic link
 >>    Craig> c. 1990 when we had precisely this problem for a few months
 >>    Craig>    -- the
 >>    Craig> number of packets in the pipe divided by the number of
 >>    Craig> connections was less than 1?
 >>
 >>    Craig> As I recall, he found some interesting behavior.
 >>
 >>Err, don't think so.  We found lots of strange behaviour in DNS, but
 >>the link wasn't over-stressed and TCP was working ok at that time,
 >>although there was a mixture of TCP implementations.
 >>
 >>Jon can correct me if my memory is fading over time.
 >>
 >>cheers
 >>ian

 cheers

   jon