FYI Eric, you may also be interested in the performance of an NT4 TCP over a
400-mile path with only 0.25% T1 frame loss -- sounds like just a little loss,
but the 1-minute transaction (~2MB transfer) took 40% longer than when no
errors occurred, all because TCP can't distinguish error loss from congestion
loss and so backs off for the wrong reason.
In the attached, the blue triangles represent ~30kB blocks from the NT box and
spaces between them indicate dead time, while TCP is timing out. The upward
movements in the red curve correspond to loss, or making no real progress
(burning daylight), while downward movements are progress at faster than the
average rate. Because of TCP's behavior, this T1 link was only fractionally
utilized. NT defaults for TCP also limit it, in terms of maximum send window
-- inadequate for this bit-rate x round-trip time, and contribute another >10%
efficiency loss.
Alex
PS: This is actually an example from a QoS paper to be presented at IEEE
WECWIS-WIP in Santa Clara next Friday.
Eric Travis wrote:
>
> >Having said all that, my understanding is that to get 'decent'
> >performance out of TCP (ie., a window > 1) packet losses should
> >be limited to less than 1 per RTT.
>
> Only if the definition of 'decent' translates to an effective
> cwnd of ~1 or 2 segments :o(
>
> You really want your (non-congestion based) losses bunched within
> single RTTs that are fairly widely spaced [how wide will depend
> on what your optimal cwnd for the path would be]. Otherwise TCP
> will be severely under driving your link.
>
> >There was a thread recently on the pilc list (see
> >http://pilc.grc.nasa.gov/pilc) on channel BER that might be
> >of interest to tcpsat participants. I am crossposting to the
> >pilc list and would like to recommend continuing this
> >discussion there.
>
> Is it me, or haven't we already seen this thread replay
> itself a half dozen times already on this *very* mailing
> list? :o)
>
> This seems like a good time to plug the excellent tcpsat
> mailing list archive that Mark makes available off of:
>
> http://tcpsat.grc.nasa.gov
>
> I'm confident that if one pulls the archive and does a
> grep on BER, errors or loss - you will be able to extract
> virtually all the collected wisdom/opinions from the cyclic
> (and often passionate) "BER Wars" thread that appear every
> 2-3 months during 1998. I'm pretty sure it is all there
> already (several times).
>
> The stuff its there for those who want to dig it out;
> Archives are there available, so let's not replay the
> old stuff again on either tcpsat or pilc? New stuff
> would be fantastic.
>
> Eric
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:54 EST