Re: [OMPI users] Disable use of Torque at run-time

[jgans]

Hi Ralph,

Thank you for you help. This is exactly what I wanted!

Regards,

Jason

Ralph Castain wrote:

Hmmm...well, here is one way to do it:

mpirun -n 1 -host n0 ./master_worker : -n N-1 -host +e ./master_worker

What this will do is put rank 0 on the first node in your allocation, 
and then all the remaining ranks on the remaining nodes in the 
allocation. All the ranks will be in the same comm_world.


Check out "man orte_hosts" for a detailed explanation (with examples) 
of this "relative node indexing" syntax.


HTH
Ralph

On Sep 10, 2009, at 3:57 PM, jgans wrote:


A single app:

mpirun -N ./master_worker

Regards,

Jason

Ralph Castain wrote:

Is the master a different app, or is the same app used?

In other words, do you run this as:

mpirun -n 1 ./master: -n N worker

or

mpirun -N ./master_worker

Either way, I can advise you on a better way to accomplish your goal

On Sep 10, 2009, at 2:58 PM, Jason D. Gans wrote:


Hi,

I have a master/worker bioinformatics application where the master 
has a

higher memory overhead than the workers. I want to restrict the master
node to a single slot (to prevent the master node from getting
oversubscribed and having workers compete for precious ram), while all
other non-master nodes can be oversubscribed (infinite max_slot).

Regards,

Jason


I guess I'm puzzled, then. First, hostfile and Torque work fine
together in the 1.3 series - it was the 1.2 series that had the 
problem.


Second, the default max_slot setting is taken from the slots 
allocated

to you by Torque. I don't see the purpose in changing them - you can
always oversubscribe the node anyway.

Perhaps you could explain more about what you are trying to do? You
may find that there is a much simpler solution already in place.


On Sep 10, 2009, at 2:07 PM, Jason D. Gans wrote:


What OMPI version are you talking about?



version 1.3.1



On Sep 10, 2009, at 1:40 PM, Jason D. Gans wrote:


Hello,

I would like to use a custom hostfile (that changes the default
max_slot
values for certain nodes). My understanding of the FAQ is that
this is
*not* possible with Torque. Therefore, is is possible to 
disable use

of
Torque at runtime (via an argument to mpirun), or do I need to
recompile
to remove Torque support altogether.

Regards,

Jason Gans
___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

Re: [OMPI users] Open-MPI and gprof

[jgans]

Hi,

Yes you can profile MPI applications by compiling with -pg. However, by 
default each process will produce an output file called "gmon.out", 
which is a problem if all processes are writing to the same global file 
system (i.e. all processes will try to write to the same file).


There is an undocumented feature of gprof that allows you to specify the 
filename for profiling output via the environment variable 
GMON_OUT_PREFIX. For example, one can set this variable in the .bashrc 
file for every node to insure unique profile filenames, i.e.:


export GMON_OUT_PREFIX='gmon.out-'`/bin/uname -n`

The filename will appear as GMON_OUT_PREFIX.pid, where pid is the 
process id on a given node (so this will work when multiple nodes are 
contained in a single host).


Regards,

Jason

Tiago Almeida wrote:

Hi,
I've never done this, but I believe that an executable compiled with 
profilling support (-pg) will generate the gmon.out file in its 
current directory, regardless of running under MPI or not. So I think 
that you'll have a gmon.out on each node and therefore you can "gprof" 
them independently.


Best regards,
Tiago Almeida
-
jody wrote:

Hi
I wanted to profile my application using gprof, and proceeded like
when profiling a normal application:
- compile everything with option -pg
- run application
- call gprof
This returns a normal-looking output, but i don't know
whether this is the data for node 0 only or accumulated for all nodes.

Does anybody have experience in profiling parallel applications?
Is there a way to have profile data for each node separately?
If not, is there another profiling tool which can?

Thank You
  Jody
___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users

  

___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users