Newsgroups: comp.parallel.pvm
From: wrankin@ee.duke.edu (William T. Rankin)
Subject: Re: can't start pvm -- BUG IN KSH???
Organization: Duke University EE Dept.; Durham, NC
Date: 1 Aug 1994 19:31:27 GMT
Message-ID: <31jiif$fbt@news.duke.edu>

In article <31bhpd$i56@acmey.gatech.edu>, ccastco@prism.gatech.edu (Costas
Malamas) writes: 

|> In article <31apdb$5bn@news.duke.edu>,
|> William T. Rankin <wrankin@ee.duke.edu> wrote:

|>  >are you sure that the PVM_DPATH variable is set correctly?
|>  >
|>  >first, kill off *all* PVM daemons on all machines.
|> 	
|>    OK, this is very weird.  After hours and hours of playing around with this
|> thing, I realized, that quite simply ksh was NOT running ANY initialization
|> files when executed from rsh.  For example:
|> rsh <remotehost> 'set'
|> would give only the basic variables -- most probably those defined by login,
|> NOT including the ENV variable that defines the ksh startup script. In other
|> words, not even /etc/profile was executed by rsh!... On this, I am positive
|> since I run it with different configurations on different machines with the
|> same results... I am in the process of transiotining to zsh, but not all the
|> machines I want to run PVM on support zsh, and quite frankly, I abhore csh.

Costas, 

I am not sure that I understand your problem.

PVM_DPATH does not need to be set on the remote machine.  It just
needs to be set *and* exported on the local machine prior running
PVM.  As is described in the Release Notes (?), PVM_DPATH overrides
the PVM_ROOT settings.  Thus it provides a workaround for the
*EXACT* situation you seem to be describing (ksh users not being
able to set environmental variables on the remote machine)

|>    Any fixes for ksh?

ksh is behaving normally.

|> Any weird, obscure back-door ideas?

Why can't you use PVM_DPATH as proscribed?  Is there a problem?
The other solution, and the one I used before the DPATH enhancement
came out, was to start PVM with a hostfile which contained the
single line:

* dx=/your/path/to/pvm3/lib/pvmd

both of these solutions are described very well in the PVM documentation.

|> (On this, I have to point out that the original reply, that claimed that
|> including PVM_DPATH on .profile worked,

(one more time)

note that PVM_DPATH is NOT needed by the remote tasks, but rather by the 
pvm daemon on the LOCAL machine.  it doesn't matter if it's in your profile
or not, it just has to be set LOCALLY before you start the PVM Master Daemon. 

|> was on an RS6000, not a standard system or OS, by any means....)

now I will usually be the first to crack on Big Blue, but in this case,
the criticism is unwarrented.  AIX is as "standard" (if there is such thing
these days) a Unix as anything SGI brings out.

as far as calling an RS-6k a "non-standard system", I'll just leave that
one alone.

Here at Duke, I support PVM on Sun-4's (SunOS 4.1.3 and Solaris 2.3),
RS-6k's under AIX, and DecStations under Ultrix.  In addition, I've worked on
networks of HP's, a KSR-1, and a Cray T3D all under PVM3.  So far, the
RS-6k's have been one of the more "well behaved" Unix flavors I've worked on.

Of course, I'm an old System-V user from my AT&T days, so your milage
may vary :-)


|> Thanx in advance
|> -- 
|> costas malamas ____________________________________________________________
|> georgia institute of technology        	    
|> OIT UA -- my opinions do not reflect those of OIT
|> internet: ccastco@prism.gatech.edu         


Good luck, 

-bill

-- 
----                                /       __/    /    /
bill rankin                        /              /    /
wrankin@ee.duke.edu               ___  /    /    /    /
philosopher/coffee-drinker       /    /    /    /    /
                                /    /    /    /    /
                             _______/  __/  __/  __/



