[SUMMARY] Can't see large (>2GB) files from SGI systems

From: David Foster <foster_at_dim.ucsd.edu>
Date: Tue May 07 2002 - 18:19:54 EDT
I could not access large (> 2GB) files on filesystems exported from
our E420R Solaris 8 (10/01) systems (latest recommended patches),
from our SGI systems.

This turned to be an SGI issue, the diversity of knowledge on
this list never ceases to amaze me!

Disabling automount and enabling autofs on the SGIs fixed the problem,
which was that automountd was reverting to NFS version 2. We in fact
had the same problem on all Sun systems, I just didn't realize it
because in my testing I was mounting explicitly, bypassing the
automounter.

One person suggested patch 111197-02, but this replaces mountd and
fixes "4389722 nfs mount of a file > 2GB is impossible", an unrelated
issue.

I'll include the following very useful reply from Andy McVey
in its entirety:

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
The symptoms you describe are typical of an NFS version 2
mount rather than version 3.  Check the /etc/fstab for manual
mounts using vers=2 as an option.  The mount command reports
the version used:

bubble 43: mount | grep ghost
jaffa:/export/ghost on /home/ghost type nfs
(vers=3,rw,intr,soft,noquota,nosuid,dev=140001)

SGIs will try vers=2 if the server doesn't respond fast
enough.  This might happen if you're using Cisco switches and
you haven't enabled the fast port spanning tree algorithm on
the SGI's port.  You'll see a lot of "server not responding"
junk in the SYSLOG as the system boots up but eventually the
mounts will succeed as version 2 mounts.  Specify vers=3 in
the fstab/automap to fix this.  One other thing - make sure
you're using autofsd and not automountd:

bubblex 44: chkconfig | grep auto
        autofs               on
        automount            off

bubblex 45: ps -ef | grep auto
  mcveya      45574      45333  0 09:20:58 pts/0   0:00 grep
auto
    root      15788          1  0   Apr 04 ?       0:02
/usr/etc/autofsd -v -m 16

automountd is old, single threaded, unmaintained by SGI and
more likely to cause tears than any other SGI supplied daemon
(I worked in the support team for SGI Europe for several
years - that's a qualified statement!)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -


In case enyone's interested, here's what I did to disable automount
and enable autofs (adjust config files for your site; assumes an NIS
source for "auto.direct"). 

Note that sometimes you have to run /usr/etc/autofs explicitly after a 
reboot to enable the new maps (this is supposed to be done automatically
on bootup).

#!/bin/csh -f
#---------------------------------------------------------------------------
# Disable automount and enable autofs on SGI systems
set hostname = `hostname`
echo "Disable automount and enable autofs for $hostname"
echo "Create /etc/auto_master"
cp /etc/auto_master /etc/auto_master.bak
cat >! /etc/auto_master << EOF
/home   auto.home       -rw,soft,intr
/-      auto.direct     -rw,soft,intr
EOF

echo "Create /etc/config/autofs.options"
cp /etc/config/autofs.options /etc/config/autofs.options.bak
set mounts = `ypcat -k auto.direct | grep $hostname | awk '{print $1}'`
set arg = "-v -m 16"
if ( $#mounts > 0 ) then
        foreach m ( $mounts )
                set arg = "$arg $m -null"
        end
endif
echo $arg >! /etc/config/autofs.options

echo "Disable automount and enable autofs"
/etc/chkconfig -f automount off
/etc/chkconfig -f autofs on

echo "Reboot system for changes to take effect."
exit 0



Thanks to:

Matt D. Harris
Andy McVey
Osama Ahmed
Weigert, Daniel
Robert Konecny

> 
> Platform: SunOS cressida 5.8 Generic_108528-12 sun4u sparc SUNW,Ultra-80
> 
> Problem: 
> 
>    On two E420R systems with recent recommended patches applied and
>    at Solaris 8 (10/01), we cannot see large (> 2GB) files on their
>    NFS mounts from our SGI systems running IRIX 6.5.14:
>    
>       $ ls -l /auto/keck1f
>       ls: /auto/keck1f/data3G.tif: File too large
> 
>    Other file utilities fail, as do basic C system calls.
>    
>    The NFS mounts are both T3 and A5100 disk arrays (only A5100's under
>    Veritas VM control), so this isn't related to VxVM. All disk
>    arrays are attached via fibre.
>    
>    I don't think this is an IRIX problem because we can see large
>    files on shares from all of our other Sun systems (U80, U60, E450).
>    What's even stranger, we can also see large files on the E420Rs'
>    system disk.
>    
>    The only difference with these two systems is that they were
>    set up for us by Sun SE personnel. The recommended patch cluster
>    installed on them is more recent than some of our other systems, so 
>    I'm stumped.
>    
> Has anyone else seen this behavior before?
> 
> Dave Foster


   << All opinions expressed are mine, not the University's >>

  =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
   David Foster    National Center for Microscopy and Imaging Research
    Programmer/Analyst     University of California, San Diego
    dfoster@ucsd.edu       Department of Neuroscience, Mail 0608
    (858) 534-7968         http://ncmir.ucsd.edu/
  =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=

   "The reasonable man adapts himself to the world; the unreasonable one
   persists in trying to adapt the world to himself.  Therefore, all progress
   depends on the unreasonable."   -- George Bernard Shaw
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Tue May 7 18:30:01 2002

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:42:42 EST