[MPICH] mpich2: smpd phrase missing? - don't share config file across PMs

[Jayesh Krishna]

 Hi,
  Are you sharing the smpd configuration file (by default ~/.smpd) across
multiple smpd daemons ? Each smpd daemon should have its own configuration
file (Only smpd root daemons modify the configuration file. Currently there
is no locking mechanism implemented on unix systems to prevent multiple smpd
root daemons from modifying the config file at the same time.).
  You could also specify the passphrase in an environment variable
(SMPD_OPTION_phrase) on each node.
  Also make sure that the environment variable "HOME" is set for all the
nodes.

Regards,
Jayesh
-----Original Message-----
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Stefan Becuwe
Sent: Tuesday, January 15, 2008 7:17 AM
To: mpich-discuss at mcs.anl.gov
Subject: [MPICH] mpich2: smpd phrase missing?


When trying to run a simple mpihello program (mpich2 1.0.6p1 / linux /
GridEngine), using f.i. 16 slots, I get the following error for some of the
nodes:

   Please specify an authentication passphrase for smpd: Aborting: unable
   to connect to node...

I've double checked that ~/.smpd contains "phrase=apassword" before starting
the job.

Since the error doesn't show up for all nodes, I assume there's some kind of
locking problem when the "dynamic_hosts" line gets updated?  Have you any
idea how to solve this problem?

Best regards


Stefan