strange server lockup - not sure how to troubleshoot

[jules]

I have a Gentoo 1.2 box exporting an NFS mount for other machines on a home network.

Whenever I move more than 50-100MBs of data across the network (no matter if it's via NFS, scp, ftp, etc.) the machine locks up, but not completely. If I go to the console, I can type a username, hit enter, it asks for password, gives me the last login time for that user, then nothing. I can also open a new ssh session to it (so at least the SSH daemon is still listening), with the same results. After the last login banner, I get no more response from the machine. If I already have an ssh session open to the machine when this happens, I can type one command (ie. ls) before I get no response.

I have no problems moving 1GB or more of data across disks or partitions on the server, either via the console or logged in via SSH.

The trigger for this condition seems to be disk AND network access at the same time.

I had no problems installing gentoo, compiling the kernel, or emerging any software (samba, etc.).

Machine specs: PII 266, 128MBs RAM, 2 Promise Ultra66 controllers (one drive per controller), Netgear NIC, Matrox mystique video.

Any suggestions for debugging techniques would be very appreciated.

Thanks.

[rac]

What kernel sources are you using, and is there anything of interest (preempt patches, low-latency, etc.) in your kernel configuration?

Is anything in the kernel logs (/var/log/kern.log maybe?) showing IDE resets or anything that looks unusual?
_________________
For every higher wall, there is a taller ladder

[jules]

Kernel: gentoo-sources

Nothing out of the ordinary in the logs. kern.log, messages, syslog & debug all clean, except for some ext3 rebuild messages immediately after the hard resets necessary to recover from the lockup.

low latency scheduling + control low latency with sysctl are ON in kernel config. Should these be off? They work fine on my desktop, though I could see why I wouldn't necessarily want them on on the server, although if I can get this bug worked out, the server will also double as the household MP3 player.

I'll try compiling the kernel without low latency and see what happens.

[rac]

I would also see if the problem persists when using vanilla sources. The gentoo-sources have lots of patches that apparently do great things for lots of people, but they seem to be responsible for a fair amount of unwanted problems, too.
_________________
For every higher wall, there is a taller ladder

[jules]

Turning off low latency in the gentoo-sources kernel didn't fix it, but moving to the vanilla-sources did.

Thanks for the tip and the fast response.

[jules]

Just a quick followup...

The problem returned when I installed a drive on the 2nd Ultra 66 controller. After much hair pulling, I checked the bios version on the cards and they were fairly outdated. Flashing the bios to the latest available version seems to have fixed the problem permanently (let's hope).