Compute node idle...

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate
star

Compute node idle...

Pupuce
Hi,

I have a problem...  i'd prefer having a dream, but... ;)

I'm trying to set up a cluster with pelicanHPC and i think it is ok because the compute nodes' screen show the welcome page.
But when i try an example and want to check it with ganglia i'm seeing nothing... the plot stay empty!!!
When i try to launch ksysguard i can configure the frontend node but i can't stay connected with the compute nodes...
I work with pelican 2.2 because i need a 32 bit version... i tried to compile a 2.5 32 bit but didn't really succeed... :(

I tryed kernel_example(...) in octave only for frontend node and it worked as in the tutorial.
then i wanted to try a parallel example with mpirun -np X octave -q --eval "kernel_example(5000, true)" it seems to work and give me some results for different X but ganglia shows nothing and htop on each node doesn't seem to show anything either...

i also change the pelican_config file and choose not tu compute with my frontend node(virtual) then the cluster bugs on the flops test because he can't comunicate with compute node!!!!

What i want is an example who works with all the compute nodes and see it somehow(ganglia and or ksysguard or else).

Sorry for my poor english i speaks much more dutch than english these last time...

Thanx in advance for your help
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate
star

Re: Compute node idle...

Michael Creel
Administrator
To get the nodes to participate, you need to specify the hostfile for mpirun, something like

mpirun -np 3 --hostfile /home/user/tmp/bhosts octave -q --eval "kernel_example(2000, true, false)"

The above noted hostfile is created when you set up the cluster (pelican_setup) but it is not optimized. You should read the Open MPI documentation to learn how to optimize the file depending on your hardware. htop will show activity on the nodes after you do this. About ganglia and ksysguard not working, I don't know what could be the problem. I haven't had any problems with ksysguard. I no longer use ganglia, it hasn't worked for a few versions now due to configuration problems.
Loading...