From: Eberhard Heuser-Hofmann [vaxinf@chclu.chemie.uni-konstanz.de]
Sent: Friday, March 17, 2000 6:22 AM
To: Info-VAX@Mvb.Saic.Com
Subject: Announcement: Message passing program TCGMSG for OpenVMS/ALPHA

Hi,

I'm proud to release the message passing environment tgcmsg 
for OpenVMS/Alpha Version 7.1, Multinet V4.2a, DEC/CPQ-C V5.7-004,
Compaq Fortran V7.2-780-4293Q.

Intro:
Many important programs for computational chemistry use message passing 
libraries in order to parallelize the code. There are several codes 
available for the message passing routines:

i.e. tcgmsg, pvm, mpi, linda.

I have access to the source of two programs that both use tcgmsg: Gamess and
Crystal98. This has motivated me to try a port of this routines.

Description:

tcgmsg consists of two parts:

1. A library called libtcgmsg.olb plus an fortran interface. 
2. A program that manages the remote execution and communication called 
   "parallel.exe". In my case I have decided to write a command-procedure
   that gets all parameters from parallel.exe and runs the parallized code
   via rshell. This little command-procedure is called child.com.

Installation:

There is no complete installation-command file.
1. create a directory i.e. [tcgmsg_vms]
1. extract tcgmsg_vms.zip
2. define a logical tcgmsg_dir as [tcgmsg_vms]
3. switch to the [.source] subdirectory
4. execute @build_tcgmsg.com
5. define a symbol para*llel:==$tcgmsg_dir:parallel.exe
6. modify HELLO.P, TEST.P and TESTF.P so that it corresponds to your cluster
   environment.
   The .p-file look like this
   <username> <ip-none-name#1> <#-processors> <executable> <scratch-dir>
   <username> <ip-none-name#2> <#-processors> <executable> <scratch-dir>
   ...
7. run tcgmsg-program
   para hello
   para test
   para testf

Important: I have noticed that in some cases the hello program has problems 
to connect to the master-program. Then the master-program stops immediately
though all rhell processes still are active. I had to stop these rshell
processes by hand.

$ pipe sys sys/clu | seac sys$input rshell

shows all bad processes in question.

I have tried to locate the problem, but I gave it up. It has something to do
with a too fast response from one node and it was possible for me to run
hello, if I changed order of the remote startup in the hello.p file. 

Remark TWO: parallel doesn't start if verify is ON!!!
   
8. switch to the [-.examples] subdirectory and produce the examples
   @MAKE_EXAMPLES.COM
   You will get eight programs and you'll find 8 .p files to start them
   accourdingly.

9. Things that have to be done:
   a. Support of UCX-sockets via socketshr.
   b. Support Multiprocessor-VMS-Machines
   c. cross platform support (Tru64<=>OpenVMS), this is what I can do here.

10. A zip-file with all parts mentioned above could be found on

    v36.chemie.uni-konstanz.de/tcgmsg_vms/tcgmsg_vms.zip

Any comments appreciated!

Eberhard

A final, more or less private remark:
I have posted here the demand of having message passing programs for OpenVMS 
about two or three times. MPI/PVM is a standard and DEQ has supported
versions for Tru64. It took me about 2 weeks until everything is
(more or less) OK in the of tcgmsg. Why is there no activity of DEQ in that
field???  

===============================================================================

Dr. Eberhard Heuser-Hofmann
Univ. Konstanz
Fakultaet fuer Chemie
Universitaets-Strasse 10
D-78464 Konstanz
Germany
Phone: +49-7531-88-2026, FAX: +49-7531-88-3139
email: vaxinf@chclu.chemie.uni-konstanz.de

===============================================================================