[Users] several questions on pugh

[Tom Goodale]

Hi,

you only need ghost-zones on internal processor boundaries.  So with 9 
procs you get 2x2x2 procs with 32 physical points each and one ghost zone 
each.

Cheers,

Tom


On Mon, 12 Feb 2007, Hee Il Kim wrote:

> Hi,
>
> I have several questions about pugh. When I run a 64 cubed single
> cpu simulation, the cactus run shows
>
> INFO (CartGrid3D): x=>[0,63]  y=>[0,63]  z=>[0,63]
> INFO (PUGH): MPI Evolution on 1 processors
> INFO (PUGH): 3-dimensional grid functions
> INFO (PUGH):   Size: 64 64 64
> INFO (PUGH):   Processor topology: 1 x 1 x 1
> INFO (PUGH):   Local load: 262144   [64 x 64 x 64]
>
> Nothing strange. But when I make a mpirun, for example, using 8 cpus (=8 cpu
> cores),
>
> INFO (CartGrid3D): x=>[0,63]  y=>[0,63]  z=>[0,63]
> INFO (PUGH): MPI Evolution on 8 processors
> INFO (PUGH): 3-dimensional grid functions
> INFO (PUGH):   Size: 64 64 64
> INFO (PUGH):   Processor topology: 2 x 2 x 2
> INFO (PUGH):   Local load: 35937   [33 x 33 x 33]
>
> Why is the local load 33x33x33? I think it should be 32x32x32 or 34x34x34
> if it includes ghostzones with stencil 1 ? I can get a correct result even
> with this problem. What does the local load represent?
>
> In order to look into further, I tried to use CCTK_MyProcs function. But I
> only get rank 0. In fact I don't know how to display the rank of computing
> node cpus. I tried "-r[o|e]" option for cactus run but no lucks.
>
> I'm using 6 node intel dual core (presler 930) cluster. I use Cactus for
> numerical relativity but I'm a newbie on computational science. I appreciate
> your kind helps in advance.
>
> Kim, Hee Il
> from Korea
> heeilkim at gmail.com
>