NFS and the beowulf cluster.

[Robert G. Brown]

 <original poster>

>
> Robert Sand <rsand at d.umn.edu>
> Sent by: beowulf-admin at beowulf.org
> 06/28/01 11:39 AM
>
>
>         To:     beowulf at beowulf.org
>         cc:
>         Subject:        NFS and the beowulf cluster.
>
>

On Thu, 28 Jun 2001 JParker at coinstar.com wrote:

> G'Day !
>
> I am not an NFS expert, nor do I play one on TV, but ....
>
> Does not the host.allow/deny file use wildcard symbols,  "  ALL:
> 192.168.*.*  ".  There is a pretty good  NFS HOWTO on LDP.

Indeed.  Currently maintained by a good friend of mine (Seth Vidal), who
gets VERY ANNOYED when folks don't read it.  He keeps a "sucker rod"
(see the syslogd man page:-) in his office to deal with folks that annoy
him.  Don't annoy him!  Check out:

  http://www.linuxdoc.org/HOWTO/NFS-HOWTO/server.html#CONFIG

for answers to nearly all your configuration questions including the
ones you haven't thought to ask yet.

There are all sorts of ways to break the exports file in subtle ways so
that it doesn't work.  And then there are netgroups.  And ipchains
(remember that RH 7.1 comes up in cosmically paranoid mode unless you
tell it otherwise, so a node might simply refuse all attempts to pass
port 2049 in either direction regardless of what daemons are running and
the state of /etc/hosts.[allow,deny]).  And chkconfig to manage all
those little symlinks.  And the fact that you need portmap running AND
accessible via ITS interaction with /etc/hosts.[allow,deny] and
ipchains.  And then there are mountd and quotad and statd.  Lots of
stuff to break.

Getting NFS running (and I may not be an "expert", but I have managed
NFS servers for about fourteen or fifteen years now:-) can sometimes be
tricky, but if you read the docs and are very careful about the format
of fstab, exports, netgroup, and so on it WILL work for you, I guarantee
it.  It isn't perfect, its performance can leaves something to be
desired, and it might even be "broken" in subtle ways if you kick it in
just the right place, but overall NFS is a really robust workhorse that
has served literally millions of users trillions of files over more than
a decade at this point.

Your best possible bet is to find somebody who "lives" in netspace near
to you (there must be any number at umn) who knows how to run NFS on
linux boxes and talk to them.  They may just look at your files and say

Oh, one more thing -- remember that a LOT of folks turn off port 2049 by
knee-jerk reflex on all their routers as NFS is NOT a secure wide area
fileservice protocol.  It looks like your $MASTER is local to your
private net and this shouldn't be a problem, but if in fact it is on the
other side of a gateway/router be absolutely certain that port 2049
isn't blocked by ipchains or router configuration.

  rgb

>
> cheers,
> Jim Parker
>
> Sailboat racing is not a matter of life and death ....  It is far more
> important than that !!!


> As reported yesterday I am still having problems with nfs on my cluster
> when I have never had nfs problems in the past. I have added the
> /etc/exports file and put all the nodes into the hosts file and added a
> line for portmap in the hosts.allow file and I still get errors on both
> sides of the equation.
>
> On the master portmap reports this when a node tries to mount the /home
> directory:
>
> portmap[3185]: connect from 192.168.1.12 to dump(): request from
> unauthorized host
>
> and in the /var/log/beowulf/node.# file I see:
>
> node_modprobe: installing kernel module: sunrpc
> node_modprobe: installing kernel module: lockd
> node_modprobe: installing kernel module: nfs
> mount: RPC: Unable to receive; errno = Connection refused
> Failed to mount 192.168.1.10:/home on /home.
>
> Now in my hosts.allow file I have "ALL: 192.168." and I have "PORTMAP:
> 192.168." and in /etc/exports I have "/home node*(rw)" I also have an
> nfs line in the /etc/beowulf/fstab that says:
>
> $MASTER:/home   /home           nfs     defaults        0 0
>
>
> What is wrong here?  I have each node listed in the /etc/hosts file too
> and still nothing.  Any ideas what might be causing this?
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu