We had nearly identical problems with our Linux boxen (various kernels 2.2.5-2.2.14) running off of our filer. We tried dozens of different NFS settings without any luck. The only thing that finally got all of our NFS problems under control was going with a custom kernel with all of trond's various NFS patches in place. RFC compliance is a wonderful thing.
Go ahead and compile new kernels with patches in place. Patches for various kernels are at http://www.fys.uio.no/~trondmy/src/. Additional information is (of course) at http://nfs.sourceforge.net/. You do not need to go with a 2.4 kernel for this; these patches can be applied (with varying levels of success) for kernels from 2.2.14 on up.
Mark
-----Original Message----- From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com]On Behalf Of Nick's Lists Sent: Thursday, April 26, 2001 8:58 AM To: toasters@mathworks.com Subject: More info on Input/Output errors...
A bit more info I forgot to include in the first place along with responses to some other folks:
We're using 2.2.16-21.7.1smp - We're holding off on 2.4 for now, as our vendor, VA Linux doens't feel comfortable with us moving up to 2.4 yet (I know, we are pushing it, but we don't want to go against their recommendations for now). If this problem is resolved in 2.4, then that'll be another thing to push us towards 2.4.
Network wise -
Transport is UDP
This is occuring on clients in multiple sites, against 7 different filers as well. I don't see how it could be a network problem tho, but I'll go over the configuration of the filers.
In all cases, the clients and servers are all on the same subnet, and in some cases, on the sames switch. I've verified that we are getting 100 mbit full duplex negotiation to the switch.
Andrew - good suggestion on the soft vs hard. I'll go over my logic for that, and try to figur eout why we're doing that. It may be ahold over from when I had a <insert name of compeditors product here> that was heavily loaded and was responding very slowly.
The thing is - other OS's aren't experiencing this problem in this environment, just the Linux boxes. That leads me to think its not a common problem like network of filer config.
Thanks!
- Nick
----- Original Message ----- From: "Traitel, Eyal" eyal@netapp.com To: "'Nick's Lists'" mrlist@noid.org Sent: Thursday, April 26, 2001 12:08 AM Subject: RE: Input/output errors from Linux clients?
Might be unrelated, and not exactly hitting your problem, but from all
we know and hear and encounter, move up to kernels 2.4 and up for real use of NFS in Linux.
What's your kernel version then ?
Eyal.
eTraitel - I'm the new eBuzzword around !!!
Eyal Traitel - Filer Escalation Engineer CNA, MCSE, CSA, LCA, NetApp CA
Network Appliance BV Holland Office Center Kruisweg 799b 2132 NG, Hoofddorp The NetherlandsOffice: +31 23 567 9685 Cellular: +31 6 5497 2568 Email: eyal@netapp.com
Get answers NOW! - NetApp On the Web - http://now.netapp.com
-----Original Message----- From: Nick's Lists [mailto:mrlist@noid.org] Sent: Thursday, April 26, 2001 8:18 AM To: toasters@mathworks.com Subject: Input/output errors from Linux clients?
I've been getting some strange errors from my Linux clients...
cp: Input/output error and cat: write Input/output error
They occur randomly when I try to copy or cat 20 - 40mb sized files to
and
from a filer. Solaris (x86 and Sparc) servers copying the same size
files
around are not experiencing the same problem.
The file systems are all mounted with nfs2, soft, bg, intr.
Has anyone else seen this before?
TIA!
- Nick