Newsgroups: comp.parallel.pvm
From: wjiang@thud.cs.utk.edu (weicheng jiang   )
Subject: Re: pvm on Paragon
Organization: Computer Science Dept, University of Tennessee, Knoxville
Date: 4 Aug 1995 18:10:17 GMT
Message-ID: <3vtnq9INNfs8@CS.UTK.EDU>

In article <3vtev4$4lk@abel.cc.sunysb.edu>, lribeiro@mathlab.sunysb.edu (Leonardo Fonseca) writes:
|> I have been trying, without success, to submit to a Paragon an
|> application parallelized using PVM.
|> 
|> The application uses the master-slave approach with both master
|> and slave programs embbeded in the same code. When using a
|> cluster of workstations, where everything works fine, the
|> code starts in one node (the workstation from where I submitted 
|> the job) which spawns the slave processes. One of the slave 
|> processes is the original workstation, which also runs the 
|> almost-no-CPU-consuming master process. The application takes 
|> two arguments, the input and output file names.
|> 
|> In order to submit the application to the Paragon, I guess
|> I have to start pvm first (e.g. "pvmd -sz xx") and then submit
|> my job using something like "?? myjob inputfile outputfile ??".
|> I guess I also have to specify where to start the job with a 
|> "-on xx" switch.
|> 

On our paragon, which requires that jobs be submitted with the script
"pexec", I have to start pvmd in one window:

xps5> pexec "pvm3/lib/pvmd" -sz 4

and then run the application in another window:

xps5> pvm3/bin/PGON/master1

If you can create your own partition on the machine, then you can start 
pvmd without pexec.

Hope this helps,
  Weicheng Jiang
  PVM Group, University of Tennessee

