[oclug] Minor network mystery
Sandy Harris
sandy at storm.ca
Fri Mar 30 00:30:47 EST 2001
This problem is solved, but I'm still wondering what caused it.
Context:
Newly installed SuSE 7.1 using 2.2.18 kernel on an old Packard Bell P233 box
that is intended to become a firewall so has two 3Com 3c905 TX net cards.
Linux sees both cards, no errors on boot.
Symptom:
Pings via one card to an NT box take huge amounts of time, anywhere from 5
to 50 seconds, mostly 10 to 25. Yes, seconds not millseconds! Pings from the
NT box to the Linux box are almost as bad. This is with only two machines on
a 100 Mbps ethernet switch, no load on either machine.
Everything on the other card side is normal. Pings take a fraction of a ms,
browsing works, ...
Failed diagnostics:
We checked BIOS IRQ settings, ifconfig, route -n. Nothing obviously wrong.
Swapping cables, trying a different port on the switch, opening up the box
and swapping the two net cards around in the PCI slots, replacing the card
that had the problem, ... No effect.
So we decided it was the machine, configured another Linux box, put it in
with same switch and cables. Works perfectly.
Methinks either that Packard Bell had some unusual ghastliness in its design
or one of its PCI slots was bad. Does anyone have more of a clue than that?
More information about the OCLUG
mailing list