[gmx-users] PARALLEL PROBLEM

[James O'Dell]

My benchmarks indicate poor scalability when using a slow
(10Mb or 100Mb) network but very good scalability when
I run on a Gigabit network.

Jim

On Fri, 2003-03-21 at 09:25, Osmany Guirola Cruz wrote:
> Hi 
>  I have a problem with gromacs and my linux cluster  look at this.
> 
> system with ~4000atoms(200ps)           system with ~40000atoms(40ps)
> 1 machine  1 micro   12589s                 1 machine  1 micro   8312s
> 1 machine  2 micro     7410s                 1 machine  2 micro   5060s
> 2 machine  4 micro     7535s                 2 machine  4 micro   5765s
> 4 machine  8 micro     6272s                 4 machine  8 micro   3701s
> 8 machine 16 micro    8783s                 8 machine 16 micro  3629s 
> i am using coulomtype=cuttoff if i  use PME its worst
> 
> and this is not very funny, I did a test with NAMD and it works 
> perfectly(scalable) but i hate NAMD, it's a real pain. which is the 
> diference? I think that my problem is the MPI library . now i am using the 
> rpm of my distribution (SuSE 8.1).(I think that i have to compile it). If 
> this is the solution ,i need to know which are the options for do it. In the 
> site of gromacs there are a rpm , is this the correct to resolve this 
> problem? ,PLEASE tell me what to do
> I have a lot of work and run a simulaton in 1 machine with the posibility to 
> use 32 it is terrible
>  HELP ME 
>             thanks
> 
> 
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.