Newsgroups: comp.parallel.pvm
From: jeff@adeno.wistar.upenn.edu (Jeffrey Taylor)
Subject: Re: PVM/Fortan & I/O problems
Organization: University of Pennsylvania
Date: 19 Dec 1996 15:18:01 GMT
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Message-ID: <59bmb9$6rr@netnews.upenn.edu>

In article <32B80A78.41C67EA6@bbn.com>,
	Denny Moore <dmoore@bbn.com> writes:
> I am using pvm 3.3.11 on a BBN TC2000 [BFLY] (don't laugh; the machine has a great
> debugger).  I am having problems with a master-slave program and file I/O for a
> Fortran test case.  I am creating a master program that spawns a worker and
> creates a direct access file for the worker to read.  The worker is apparently
> opening the file OK, however, I get a file read error (past end-of-file) as if
> the file doesn't exist.  I suspect the worker is not getting the correct file
> handle info.  Also stdout messages from the worker are not being printed.  Something
> is wrong with file I/O; what am I missing.
> 
> I have attached the master & worker fortran src, the make file and a print of
> the messages that I get when I run the sample.  
> 
> Any help would be greatly appreciated.
> 
> 

I don't usually debug opc but since it's almost Christmas...

Your slaves stdout messages are not supposed to be printed.  RTFM.  They end up
in a log file somewhere.  If you want them to be printed, call pvmfcatchout()
from the master.

The file io code *looks* ok, but why would you want to do this anyhow?  Why not
just pack the data into a buffer and pvmfsend it to the slave?  It may be that
the O/S is caching the file somewhere that the slaves don't see.  What if the
file exists before you start the master (i.e. don't have the master generate
the file)?  Can the slaves read it then?  If not, they are probably looking in
the wrong place.  This may sound absurd to a fortran programmer, but using file
names, with complete paths, can be very helpful with this sort of thing.

Jeff Taylor


