How to best utilize supercomputer?

[Srihari Sritharan]

Hi guys,

So I just compiled Hermes on my school's supercomputer, and I'm trying
to solve Schrodinger's equation using my adaptive solver, basically at
a rate faster (and more memory) than I could achieve on my personal
laptop. I have a few questions regarding it's execution.

I used the MPI execution of Hermes, only to realize that this opens
multiple instances of the program, which all communicate with each
other.
First question: why is this beneficial? i.e. why do I want to have
multiple copies of hermes running in parallel, with MPI communicating
between the instances?

Secondly, I would prefer to run one instance, utilizing as much memory
and as many CPU cores as possible; however my supercomputer
administrator informed me that this is not possible unless Hermes was
built from the ground up with this intention in mind. When you guys
have access to a supercomputer, how do you usually utilize it to
maximize the computation that hermes3d calculates?

Thanks, I'm just a novice to supercomputers in general.
Sri

[Lukas Korous]

Hi,

the bad thing is that Hermes is not a parallel library. This means that
you will always use one core (with as much memory you have). We plan to
change this in future.

Lukas

[Srihari Sritharan]

Ah, dang. Well thanks, I'll check in the future for such changes!

Sri

[Pavel Solin]

On Fri, Jul 1, 2011 at 5:08 PM, Lukas Korous <lukas....@gmail.com> wrote:

Hi,

the bad thing is that Hermes is not a parallel library. This means that you will always use one core (with as much memory you have). We plan to change this in future.

Lukas, are you sure about Hermes3D? I know that David did such 

experiments before he left, but I am not sure whether this is in 

upstream or not. 

Pavel

 

Lukas

On 7/1/2011 9:50 PM, Srihari Sritharan wrote:

Hi guys,

So I just compiled Hermes on my school's supercomputer, and I'm trying
to solve Schrodinger's equation using my adaptive solver, basically at
a rate faster (and more memory) than I could achieve on my personal
laptop. I have a few questions regarding it's execution.

I used the MPI execution of Hermes, only to realize that this opens
multiple instances of the program, which all communicate with each
other.
First question: why is this beneficial? i.e. why do I want to have
multiple copies of hermes running in parallel, with MPI communicating
between the instances?

Secondly, I would prefer to run one instance, utilizing as much memory
and as many CPU cores as possible; however my supercomputer
administrator informed me that this is not possible unless Hermes was
built from the ground up with this intention in mind. When you guys
have access to a supercomputer, how do you usually utilize it to
maximize the computation that hermes3d calculates?

Thanks, I'm just a novice to supercomputers in general.
Sri

--
hermes3d mailing list
herm...@googlegroups.com
http://groups.google.com/group/hermes3d?hl=en

--
Pavel Solin
University of Nevada, Reno
Home page: http://hpfem.org/~pavel
Hermes: http://hpfem.org/
FEMhub: http://femhub.org/

[Lukas Korous]

Hi,

I am sure. I had checked to be sure.

Lukas