Leo Bicknell wrote Your problem is right here: 14 urish-pitts-gw.centrepc.net (67.97.250.166) 100 msec * 100 msec NT in many configs defaults to a TCP Window size of 8k, other NT and Server 2000 default to 16k.
This seems premature as a conclusion to me. If they don't have issues with Japan where the best RTT they can get is around 130ms I don't see why 100ms would be an issue. Michel.
In a message written on Thu, Mar 18, 2004 at 11:08:21AM -0800, Michel Py wrote:
Leo Bicknell wrote Your problem is right here: 14 urish-pitts-gw.centrepc.net (67.97.250.166) 100 msec * 100 msec NT in many configs defaults to a TCP Window size of 8k, other NT and Server 2000 default to 16k.
This seems premature as a conclusion to me. If they don't have issues with Japan where the best RTT they can get is around 130ms I don't see why 100ms would be an issue.
Because these are theoretical maximums. Very minor other issues (eg some packet reordering, 5-10ms of jitter, a single lost packet every now and then) will easily cut this in half. 130ms link at 95% of max (about the best you do on a clean link) would be 59k/sec, a 100ms link with 10ms jitter would drop to around 50% of max, or 40k/sec, or 33% slower. I bet setting the windows to 32k on both sides fixes his problem. -- Leo Bicknell - bicknell@ufp.org - CCIE 3440 PGP keys at http://www.ufp.org/~bicknell/ Read TMBG List - tmbg-list-request@tmbg.org, www.tmbg.org
participants (2)
-
Leo Bicknell -
Michel Py