[wplug] NFS: Copying large files hangs

Michael Skowvron michaels at penguincentral.org
Mon Sep 12 12:06:12 EDT 2005


At this point, it could be a hardware or software problem. Does the 
problem happen with _any_ large file, or just one particular large 
file? Some hardware problems are only exposed by particular data 
patterns.

Assuming the problem effects all files, first determine if it's 
related to the NFSv3 code path by switching to NFS version 2 and 
performing the same test. If NFSv2 fails the same way, the next step 
would be protocol analysis. Repeat the test while capturing the entire 
session with ethereal. Capture on both the client and the server. 
Capture the first 300 bytes of all packets.

Ethereal should be able to show you where the transfer starts to have 
problems. Comparing the server and client traces may show where the 
problem is. If you make the capture files available (gzipped), I'll 
take a look at them.

It is extrememly important to ensure that _no_ packets are dropped by 
ethereal during the capture. You won't need promiscuous mode. Don't 
update list in real time. Disable MAC, network, and transport resolution.

Michael



Brandon Kuczenski wrote:

> I am having some issues with NFS.  The Client machine is Linux 2.6.11, 
> running Debian, and the server is FreeBSD 4.11.  When I copy a large 
> file (~200 MB) I get about 3/4 of the way through before the copying 
> just stops.  I can't tell how far it gets but it's not the same point 
> every time.
> 
> Before the glitch occurs, ping times between the machines are around 
> 1ms; afterwards, they vary from 500-1500ms.  The client machine becomes 
> unstable and sometimes won't even shutdown cleanly (I get an endless 
> screen of illegible nfs/RPC error messages) but it is fully repeatable.
> 
> I believe both client and server are running NVS version 3.  Here is the 
> line from /etc/fstab on the client:
> ocean:/home/share /mnt/nfs/ocean/share  nfs 
> nfsvers=3,rsize=8192,wsize=8192,timeo=14,intr   0       0
> 
> I added all those options after running with defaults failed.
> 
> Any suggestions how to solve this problem? or at least how to approach it?
> 
> -Brandon
> 
> _______________________________________________
> wplug mailing list
> wplug at wplug.org
> http://www.wplug.org/mailman/listinfo/wplug



More information about the wplug mailing list