forcedeth issue? [SOLVED]

[sleepy boy]

hi,

I've got a FreeBSD server and and I daily upload a few gigs of backups to it.

Whenever I start the uploads from my (2 weeks old) gentoo system, to the nfs mounts of the fileserver, I'm getting strange problems after a few minutes of uploading.
The upload seems to stall.(no network data in or out of my system) but when I start to check, the strangest things happen. I can't kill processes(like the backup moves that are stalling) the mysql client seems to lose all connections with its database. And If I try to reboot after all this happened, NOTHING happens, except the warning messages on open consoles. Sometimes the system also hard-locks...

just for info: ssh connections to the same server stay open
I've checked my logs but nothing related is in there.

The system is:
amd x2 3800
msi k8n neo2-fx (nforce3 chipset)
agp nvidia 6600GT
1.5Gb Ram
WD 250Gb sata1 HD 7200RPM

I'm thinking this is a forcedeth bug or something. but I'm not sure.

Is there anyone with a clue about what is happening???
I've been researching this problem for a week now and I'm getting hopeless

Tnx in advance

Sleepy boy

[Raffi]

You don't say which kernel you are running. I have seen many problems with the forcedeth driver. You can try a later kernel (or maybe just a different one).

[James Wells]

Greetings,

Whenever I start the uploads from my (2 weeks old) gentoo system, to the nfs mounts of the fileserver, I'm getting strange problems after a few minutes of uploading.
The upload seems to stall.(no network data in or out of my system) but when I start to check, the strangest things happen.

A couple things I would suggest looking at;

• Run iostat 2 on the server while performing an upload -- I suspect you are running into an IO overload on the NFS server. Once this happens, your client will lose connectivity to the NFS server and will block unless you have in configured in soft mode.
• Run nfsstat on the client while performing an upload -- Check for NFS errors during the transfer.
• Check your NFS mount options -- Ensure that your NFS mount is configured for soft mode, SYNC. SYNC will be slower than ASYNC, but it will give your server time to clear the buffers every few seconds.
• Check memory utilization on the server and client during a transfer. If you are getting the NFS errors or the high IO listed above, you will most likely see the memory usage on the server start going through the roof if the issue is on the server side or extremely high cache usage on the client if it is a client issue.

[sleepy boy]

tnx for the replies,

as a first, I'm running the stock gentoo 2.6.17-r5 kernel


I've changed my nfs mount line to this:
192.168.1.3:/backups /mnt/backups nfs rw,sync,soft 0 0

And now it seems to be running stable,
but instead of the backups taking 20 minutes it's going to take around 2hours now.
The upload speed dropped from 6700kb/s to 1300kb/s approx. So is there any way to (safely) speed this up?

iostat: output was pretty calm, the server was still 60% idle and the hd throughput on the drive was around 10MB/s with still plenty of memory free
nfsstat didn't point to anything significant imho.

Thank's for the tips about nfs, they really helped

[Raffi]

Using soft on nfs is probably not a great idea. At the very least, you will want to check the integrity of your backups. If soft sees an error, it will go on with your file corrupted.

Try an older kernel to see if the network driver is more stable. You can also try getting the latest nvidia net driver from nvidia and seeing if it works.

[sleepy boy]

On that I agree, I've been looking around at the different mount options for nfs filesystems, and now i'm going with:

Code: Select all

192.168.1.3:/data       /mnt/data       nfs             rw,sync         0 0

What I have noticed is when I ran with async, nfsstat showed thousands of retransmits, but on the server there didn't seem anything wrong...

so I'll look into that ...

Tnx for the help,

greets