Date: Sun, 16 Mar 2003 20:50:47 +0100 From: "Olivier Dony" <odony@student.info.ucl.ac.be> To: "Bill Moran" <wmoran@potentialtech.com> Cc: "Simon Barner" <barner@in.tum.de>, <freebsd-questions@freebsd.org> Subject: Re: Too many collisions on network? Message-ID: <025601c2ebf5$5abc25f0$1502a8c0@blacktrap.net> References: <005d01c2ebcb$82b343b0$1502a8c0@blacktrap.net> <20030316152335.GA1434@zi025.glhnet.mhn.de> <002201c2ebd3$92d0a7d0$1502a8c0@blacktrap.net> <3E74B8BC.4030009@potentialtech.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sunday, 16 March, 2003 18:47, Bill Moran wrote:
> I have seen cases where the media type was auto-negiotiated wrong. I think
> this was due to crappy wiring, but the circumstances didn't allow for a lot
> of experimenting.
FWIW : On the 3 servers where I have tested this, autoselect had every time
selected half-duplex mode, and it seems a wrong choice for at least 2 of those
3 servers, see below.
> I would suggest talking to your ISP and verifying what the connection _should_
> be. Then work towards getting it there. When the two ends see something
> different, performance blows.
I will try to contact the ISP during business hours and see what they
tell me, and ask about this kind of shortage that happened too, because
I still haven't figured out if the problem was network or server-related.
> Going to full-duplex should reduce collisions to 0, and give you the max
> performance available. I just did some experimenting with turning duplex
> from half to full and back on my computer here, and if there's any interruption,
> it was less than I could easily measure. Don't know if that'll be the same
> with all switches or not.
Well after doing some poking around on other servers first, I did change the
mode and happily that didn't interrupt the connection. Also the collisions
did drop to 0 on all 3 servers where I tried it, with a small difference :
o On the other 2 servers I changed from autoselect 10BaseT/UTP h-d
(resp. 100BaseTX/UTP) to 10BaseT/UTP full-duplex (resp. 100BaseTX/UTP f-d),
and collisions dropped to 0, nothing else seems to have changed, still no
i/o errors, and no change in the bytes throughput.
o On the server I was talking about earlier, here is an excerpt of netstat
while switching from half to full-duplex and back :
input (Total) output
packets errs bytes packets errs bytes colls
900 0 141953 1261 0 1351094 633
938 0 163593 1048 0 1157276 543
782 0 126538 938 0 1090243 414
771 0 124376 987 0 1217638 493
894 0 161036 1059 0 1111060 573
913 0 123942 1092 0 1028476 562
-> going full-duplex... here I guess
601 11 72155 695 0 569992 132
461 8 61467 566 0 560226 0
462 12 61552 546 0 649187 0
477 6 72589 555 0 649629 0
517 13 79602 624 0 668592 0
-> back to half-duplex with autoselect .. looks like here
0 0 42814 0 0 500765 0
815 0 150286 1128 0 1841082 749
1062 0 176522 1478 0 2051510 554
The input errors and other numbers were consistent during the few minutes of
testing in all 3 cases, but I cut it to a few lines for the sake of the
mailing-list. Is this increase in input errors and drop in bytes throughput a
problem? I guess the input errors are not good, when we can see that there are
no real erroneous packets coming in before. As for the change in throughput
while the server load stayed constant, it doesn't look good either, and I didn't
notice the same behaviour on the 2 others. Those don't have the same load though.
Any ideas? And thanks again, I've learned a lot so far with your kind help :-)
Olivier
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?025601c2ebf5$5abc25f0$1502a8c0>