Packet corruption in re0 (and em0 too)
raul at pop.isdefe.es
Fri Feb 22 13:56:59 UTC 2008
Christoph Schug escribió:
>> The fastest method I found to reproduce the problem is receive a
>> compresed tar (with good compression ratio) over the network (using nc,
>> 100 Mbps fdx), pipe through tar that decompress and write it on disk (a
>> lot of MB/s and tps looking at 'systat -vm'). The source tar file is
>> about 5Gb and link always fails before EOF :|.
> Up to now, no problems, but I certainly need more uptime to be sure.
> Maybe I can tell more later that day.
My box used to also need a couple or three days to exhibit the problem
until I realize that my previous example made it happen in minutes:
RELENG_7_0 i386 - em0 - dot1q - switch - clear - re0 - RELENG_7_0 amd64
on the i386: cat files_with_lots_of_zeros.tgz | nc -l 3000
on the amd64: nc i386_box_address 3000 | tar xzvf -
My files_with_lots_of_zeros.tgz is about 5Gb, and the i386 box flood the
re0 link (1000baseTX vs 100baseTX). I've never seen the end of the file
on the re0 side.
No problem on the i386 side, just to illustrate my setup.
I see the same behavior on the amd64 box using an em0 card (pci one). No
clue if my problem differs from yours but in my case it doesn't look
like only a re0 problem :/.
More information about the freebsd-current