Newsgroups: comp.parallel.pvm
From: pefv700@chpc.utexas.edu (Christopher Phillips)
Subject: Re: PVM is not an Operating System (was: pvmfs )
Organization: The University of Texas at Austin, HPCF
Date: 29 Sep 1994 12:06:51 -0500
Message-ID: <36es7b$b1v@mintaka.cc.utexas.edu>

In article <36cj53$p2c@falcon.ccs.uwo.ca>,
Crispin Cowan <crispin@csd.uwo.ca> wrote:
>In article <sssfagg.780241446@reading>,
>Graham Edward Fagg <sssfagg@reading.ac.uk> wrote:
>>	can anybody tell me if the pvm fs does proper file locking for
>>	multiple treads accessing a common file?
>
>There is no "pvmfs".  PVM uses the host's file system (usually UNIX) so
>you get whatever file locking semantics are provided by the host OS.
>
>[...] Except for message
>passing, all other services are provided by the host operating system.

Well actually there is a pvmfs.  Look for pvm334fs.patch in
pvm3/contrib/pvmfs on ftp.netlib.org.  This file is a patch and should
be applied to the 3.3.4 source.  Unfortunately, there is currently no
support for file locking.  From Readme.fs:

    PVMFS is a PVM task process which facilitates I/O for programs
    utilizing the Parallel Virtual Machine system (see pvm(1PVM)) on a
    cluster of workstations.  The file server presents a common file
    system and support for coordinated I/O to its parent and siblings.

This is all done by having an initial parentless task fork and exec
pvmfs.  pvmds give all tasks a fstid which all analogous to ptids.
(Most) I/O system calls are intercepted and request I/O services
from the pvmfs.

Chris

