[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: #4: Re: [ptp-user] How do I set up a machine set?
|
Tudor,
See comments below.
On Aug 16, 2007, at 1:13 PM, Tudor Buican wrote:
Greg,
Here are a couple of observations:
1. The nodes were initially on "idle" (allocated to me, etc.), then
they
turned to "job running" when I started a job, and to "job stopped" at
the end of the job. This worked fine, and I could see which nodes the
job was running on.
However, when I turned off precisely the nodes the job had run on and
restarted that same job, not only did those nodes not change their
status from "job stopped" to "down" (nor to anything else), but the
job
itself (the icons) showed up as "exited normally", even though the
"Process details" window showed "Node: N/A" and no program output.
I suspect that we're not dealing with this situation correctly (it
may never have been tested in fact). Would you please open a bug
report so we can keep track of the problem and make sure it gets fixed?
2. The MPI Artifacts/Run MPI Analysis doesn't seem to find any such
artifacts in my MPI program.
Beth can probably answer this better than me, but the problem is
usually that you haven't set the location of the MPI header files in
the project properties.
Apart from these teething problems (and whatever else may be
lurking out
there), this is very nice work!
Thanks for the feedback and compliments!
-Tudor
On Thu, 2007-08-16 at 12:13 -0400, Greg Watson wrote:
Tudor,
Has the node icon changed to 'down'? If so, then it sounds like it
may be a bug.
Otherwise, the node state, and consequently what is displayed in the
view, depends on the events OMPI sends, so it's possible that OMPI is
not detecting the node is down. There are definitely some issues with
the way OMPI detects these kinds of things.
Greg
On Aug 16, 2007, at 11:09 AM, Tudor Buican wrote:
Greg,
I have OpenMPI 1.2 running, with devel-headers, etc. PTP seems to be
running nicely, and I was able to build and run parallel jobs on my
cluster through it. It also shows the machines and the jobs on the
cluster.
One quick question, though: The "Node Info" display doesn't seem
to get
updated and it claims that nodes that I had turned off in the
meantime
are still "up". Is this how it should be?
Thanks again,
-Tudor
On Thu, 2007-08-09 at 17:14 -0400, Greg Watson wrote:
The quick answer is I don't know, but it would be worth a try
though.
Once you've installed the RPM, check in the <install_dir>/include/
openmpi directory. If there are sub-directories opal, orte and ompi
that have lots of header files in them, then you're probably ok.
You
could then try re-running the BUILD script an make sure that the
build goes ok. Let me know if you have problems.
Greg
On Aug 9, 2007, at 5:06 PM, Tudor Buican wrote:
Greg,
I have one quick question: if I install the rpm's openmpi-runtime
and
openmpi-devel, would that be equivalent to building and installing
openmpi from the tarball with the -with-devel-headers option?
Thanks again,
-Tudor
On Thu, 2007-08-09 at 10:25 -0400, Greg Watson wrote:
Tudor,
Did you install OpenMPI yourself, or are you using a pre-
installed
version? The fact that 'openMPI is all over the place' makes me
suspect you're using a pre-installed version. PTP only works with
version 1.2 of OpenMPI, so that's the first thing to check. When
OpenMPI is built for your platform, it *must* have been
configured
with the '--with-devel-headers' flag or it will not work with
PTP.
This is documented in the PTP release notes here: http://
www.eclipse.org/ptp/docs/releases/release-1.1.php. The PTP
manual is
available from the Eclipse built-in help, or here: http://
www.eclipse.org/ptp/doc.php.
Greg
On Aug 9, 2007, at 9:55 AM, Tudor Buican wrote:
Good morning.
As it turns out, openmpi-default-hostfile was hiding in plain
sight
in /etc. I put in the node names, etc., and mpirun, for example,
sees
them and uses them by default. However, Eclipse/PTP still
doesn't, and
displays only one node with no information.
Here's what shows up on the console when I start eclipse (in
the PTP
Runtime perspective):
............................................
[tudor@herd-0 ~]$ eclipse&
[tudor@herd-0 ~]$ In getResourceManagerFactories
retrieved factory: Simulation,
org.eclipse.ptp.simulation.core.resourcemanager
leaving getResourceManagerFactories
Your Control System Choice: '<UNDEFINED CS>'
Your Monitoring System Choice: '<UNDEFINED MS>'
No previous (or invalid) control or monitoring system selected.
Default systems set to Open Runtime Environment (ORTE). To
change,
use
the Window->Preferences->PTP preferences page.
Your Default Control System Choice: 'Open Runtime Environment
(ORTE)'
Your Default Monitoring System Choice: 'Open Runtime Environment
(ORTE)'
In retrieveConfigurationWizardPageFactories
wizard page factory:
org.eclipse.ptp.simulation.ui.wizards.SimulationRMConfigurationW
iz
ar
dP
ageFactory@39e2ee3f for class: class
org.eclipse.ptp.rmsystem.SimulationResourceManagerFactory
leaving retrieveConfigurationWizardPageFactories
XXXXXXXXXXX refreshRuntimeSystems(false), isInitialized():false
XXXXXXXXXXX refreshRuntimeSystems calling initialize(),
force:false,
isInitialized():false
refreshRuntimeSystems
SHUTTING DOWN CONTROL/MONITORING/PROXY systems where appropriate
User selected 1 simulated machines.
SimPreferences: Machine 0 = 1 nodes
ModelManager.runtimeNodeGeneralName - #keys = 2, #values = 2
Unknown machine ID (machine0), adding to the
model.
Unknown node number (0), adding to the model.
...........................................
Do you have any idea what it is that I'm doing wrong?
Thank you so much for the help.
-Tudor
On Wed, 2007-08-08 at 14:12 -0400, Beth Tibbitts wrote:
The hostfile you want is located here:
<openmpi-install-dir> e.g. /usr/local -- wherever you installed
OpenMPI
<openmpi-install-dir>/etc/openmpi-default-hostfile
...Beth
Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools http://eclipse.org/
ptp
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 455 Park Place, Lexington, KY
40511
Tudor Buican
<tudor@xxxxxxxx>
Sent
by: To
ptp-user-bounces@ PTP User list
eclipse.org <ptp-user@xxxxxxxxxxx>
cc
08/08/2007 01:26
Subject
PM Re: [ptp-user] How do I
set
up a
machine set?
Please respond to
PTP User list
<ptp-user@eclipse
.org>
Thank you. All I have is a file I created in my home directory
which
lists the machines in the cluster and which I give as a
parameter
to the
-machinefile option when I call, say, mpirun under openMPI. Is
there any
way of pointing PTP to that file? I can't find any file called
openmpi-
default-hostfile, nor any "etc" directory associated with the
openMPI
installation (I assume that's what Greg had in mind).
Also, is there any document that could be used as a PTP
manual? I
have
in mind something covering small but essential issues such as
this
one.
Thanks again,
-Tudor
On Wed, 2007-08-08 at 12:51 -0400, Beth Tibbitts wrote:
See this:
http://dev.eclipse.org/mhonarc/lists/ptp-user/msg00192.html
...Beth
Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools http://
eclipse.org/ptp
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 455 Park Place, Lexington, KY
40511
Tudor Buican
<tudor@xxxxxxxx>
Sent by:
To
ptp-user-bounces@ ptp-user@xxxxxxxxxxx
eclipse.org
cc
Subject
08/08/2007 12:44 [ptp-user] How do I set
up a
PM machine set?
Please respond to
PTP User list
<ptp-user@eclipse
.org>
Hi.
I've just installed Eclipse and PTP and I can get the PTP
Runtime
perspective. However, the "Machines" view only shows one
machine
with a
question mark on it, and I can't find any way of setting up a
set of
machines. Nor can I find a manual for PTP.
I'm running Eclipse/PTP on the head node of a five-node Rocks
4.3
cluster, with openMPI enabled by default.
Any help would be greatly appreciated.
Thank you.
-Tudor
--
Tudor N. Buican, PhD
Managing Member
SEA LLC
phone: 505-271-9925, 505-818-7303
fax: 505-271-9925
email: tudor@xxxxxxxx
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
--
Tudor Buican <tudor@xxxxxxxx>
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
--
Tudor N. Buican, PhD
Managing Member
SEA LLC
phone: 505-271-9925, 505-818-7303
fax: 505-271-9925
email: tudor@xxxxxxxx
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
--
Tudor Buican <tudor@xxxxxxxx>
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
--
Tudor Buican <tudor@xxxxxxxx>
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
--
Tudor Buican <tudor@xxxxxxxx>
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
