[Sheflug] NFS Issues, Alternatives?

[James Wallbank <james [at] xxxxxxxxxxx>]

Hello All,

At Access Space a problem seems to be emerging with our NFS server.

For a few years the NIS/YP and NFS system we've had working has been 
working really well, with a 500mHz CPU with 512mB RAM running a 
dedicated IDE disk to serve everyone's home directories.

However, recently we've found that it's possible to lock up the server 
simply by moving large numbers of files in a single operation. We've 
locked up the server with GFTP, by backing up websites containing 
thousands of files (interestingly, wget doesn't prompt the same problem) 
but more disturbingly, recently I dragged and dropped a directory 
containing 3400 files and locked it up instantly.

What happens is the machine completely locks up. You can ssh in to it, 
but after it challenges you for a password, you never get a command 
prompt - clearly it's very, very busy doing something else. The only 
solution has been to give it the three-fingered salute from the keyboard 
attached to the server. But even then, the machine starts to shut down, 
gets to the step "Stopping NFS Services" and hangs. Then I have to power 
cycle it to get it up again. Ouch!

I haven't found any meaningful clues in /var/log/messages or /var/log/syslog

The machine is running Mandrake 9, and I have heard rumours of an 
rpc.lockd bug on that version of the distro, so I've updated the 
nfs-utils package in the hope that this will cure it for now. Failing 
that, I'm intending to uprade to Mandrake 10, and possibly upgrade the 
hardware too.

I run two other NFS servers and both of them are running newer versions 
of Mandrake, and they haven't had these problems, so I am hopeful. Then 
again, they (the other servers) are less heavily used.

However, LOTS of people say that NFS is always problematic. I wonder 
whether any of you have implemented an alternative exported filesystem 
that makes an ext3 device available to client machines in a totally 
transparent way, preserving file permissions like NFS. As I understand 
it (I haven't implemented it) Samba requires another layer of passwords, 
and doesn't behave _exactly_ like a local drive, whereas NFS does 
(except when it locks up!). Is this corect, or could I easily implement 
Samba as an alternative exported FS?

Of course, it's always possible that we have a hardware issue. Is an IDE 
device simply not up to handling these sorts of many-thousands of reads 
and writes in a single operation? Should we upgrade to SCSI or a RAID array?

Your thoughts welcome.

Thanks,

James
=====
___________________________________________________________________

Sheffield Linux User's Group -
http://www.sheflug.co.uk/mailfaq.html

 GNU the choice of a complete generation.