[comp.sys.sgi] REPOST - problems with pcnfs

robert%zeus@swanee.ee.uwa.oz.au (Roberto Togneri) (09/11/90)

About a week ago I posted a news article regarding a problem we are
experiencing with running pcnfs and mounting filesystems from our
IRISes. I think somebody from Silicon Graphics tried to mail me a reply
but there was a bug in our setup which only left a remnant of that mail.
This bug has been fixed (I've got replies to another article I posted)
so I'll repeat my problem here:


Our setup consists of three IRISes running 3.1d and NFS. Connected
to these are PC's running PC-NFS 3.0.1. The IRISes are cross-mounted
via NFS and the PC's mount each root filesystem of the IRISes on
separate drive letters. A peculiar problem has always existed which
I don't recall has been  mentioned in this newsgroup. The problem 
occurs when attempting to copy or even do a directory listing of the
mounted IRIS filesystems on the PC's. Every now and then the copy
command will hang on a particular file. The rcp will sometimes work
on this file but will sometimes hang as well. Doing ls or dir of
some IRIS directories will also hang. The problem in this case usually
seems to be the number of files on the directory; removing a file or
creating one solves the problem! Needless to say this also creates 
problems when using the IRIS printer from PC-NFS. The net print command
will hang on some files because they can't be copied to the spool area.

Has anybody come across these problems? 

Another problem which I think has been covered is the incompatibility
of NFS between IRISes and Suns. We can mount a Sun filesystem happily
enough but the Suns can't mount the IRIS filesystems. The mount
program returns a version mismatch diagnostic.  Is this correct?
Any solutions? Is this problem related to the above problem?

Any help in this matter would be greatly appreciated.
--
Dr. Roberto Togneri
Dept. of EE Engineering                 ACSnet: robert@swanee.ee.uwa.oz
The University of Western Australia     INTERnet: robert@swanee.ee.uwa.oz.au

bernie@umbc5.umbc.edu (Bernard J. Duffy) (09/19/90)

In article <robert.653010246@zeus> robert%zeus@swanee.ee.uwa.oz.au (Roberto Togneri) writes:
>
... 1st portion ... deleted ...
>
>Our setup consists of three IRISes running 3.1d and NFS. Connected
>to these are PC's running PC-NFS 3.0.1. The IRISes are cross-mounted
>via NFS and the PC's mount each root filesystem of the IRISes on
... delete text (rest of this question)...

     I've got a 4D/25 with 3.2 Irix and PC-NFS (version unknown) and
that's been working for some time.  The PC-NFS came as part of the
FTP Software (FTP is a company name).  The FTP software was easy
to setup and it uses the NameServer BIND name-resolution routines.  I
know nothing more about this setup beyond these comments.

>
>Another problem which I think has been covered is the incompatibility
>of NFS between IRISes and Suns. We can mount a Sun filesystem happily
>enough but the Suns can't mount the IRIS filesystems. The mount
>program returns a version mismatch diagnostic.  Is this correct?
>Any solutions? Is this problem related to the above problem?

    I ran into a similar problem with mounting NFS exported directories
from one machine to another.  The problem was the "hostname" that was
being sent in the mount request was the short "one word" (titan1) and
the remote system needed that plus the next sub-domain piece (titan1.math
of titan1.math.umbc.edu).  The remote system was "umbc5.umbc.edu" and
it added ".umbc.edu" to the end of  titan1 and got titan1.umbc.edu which
wasn't a valid node.  This particular problem was from {Sta,A}rdent's
titan 3000 machine, but I would expect that this could happen to any
node that only uses 1 part names for their hostname (`hostname`) which
the SGI's love to do when using /etc/hosts name resolution.  The fix
came from changing the software set "hostname" (/etc/init.d/tcp or
the like) so it used the two part name (hostname titan1.math).  I
actually did it on that Titan machine by changing the NVRAM entry which
would be similar to changing the Prom Monitor's hostname environment
variable.  I HAVEN'T DONE THIS on an SGI.  Much of this single name
(the "only SGIs on the network" assumption) problems may have completely
disappeared with the name service setups.... I haven't converted our
current SGI's to that setup.

     This is really a general comment on NFS mounts with other machines
that have different versions of NFS running.

>
>Any help in this matter would be greatly appreciated.
>--
>Dr. Roberto Togneri
>Dept. of EE Engineering                 ACSnet: robert@swanee.ee.uwa.oz
>The University of Western Australia     INTERnet: robert@swanee.ee.uwa.oz.au

     The IRIX 3.2's version of NFS seems to work as expected on these
and "showmount -e" commands.  My particular problem was with another
machine type.

     Hopefully all venders will catch up with the "best" NFS protocols
     and features.

Bernie Duffy   Systems Programmer II | Bitnet    :  BERNIE@UMBC2
Academic Computing - L005e           | Internet  :  BERNIE@UMBC2.UMBC.EDU
Univ. of Maryland Baltimore County   | UUCP      :  ...!uunet!umbc3!bernie
Baltimore, MD  21228   (U.S.A.)      | W: (301) 455-3231  H: (301) 744-2954