[MITgcm-support] installing problem

[Martin Losch]

Hi Ali,

that depends entirely on your computational domain and your available resources.

As a rule of thumb I make sure that the horizontal tiles do not become smaller than say, 30x30 (sNx, sNy) grid points. Below that number the global pressure solver (cg2d) usually does not scale linearily (that does not mean, that the code will not get faster if you have 2 30x15 instead of one 30x30 tile, but it will most likely not be 2 times faster). With multicore cpus it is possible (and likely) that the bandwidth of the core connection to the memory bus is the limiting factor. In the past I had situations with intel quad-core cpus where using 4cores on one cpu/chip was no faster than just using 2 per cpu/chip. in the end you'll have to figure out what is most efficient for your particular application.

Martin

On Mar 4, 2011, at 7:43 PM, Ali Nazemian wrote:

> Sorry, about size.h accutally how i should assign processors? we have
> 180 available core on our cluster, i should set size.h for using all
> 180 cores or something else?
> Regard.
> 
> On 3/4/11, Ali Nazemian <alinazemian at gmail.com> wrote:
>> Lol.
>> Thanks.
>> 
>> On 3/4/11, Martin Losch <Martin.Losch at awi.de> wrote:
>>> Hi Ali,
>>> 
>>> I suggest that you read the error message(s) carefully. I have removed
>>> the
>>> lines in between the actually error messages from the compiler below, to
>>> make it a little easier (-:
>>> On Mar 4, 2011, at 1:56 PM, Ali Nazemian wrote:
>>> 
>>>> Hi,
>>>> I am using to install MITgcm on Rocks Cluster (Centos 5.4x86_64)
>>>> genmake2 --mpi and make depend command going well and without any
>>>> error. but when i submit make command, some errors showed up:
>>>> 
>>>> mpif77 -fconvert=big-endian -fimplicit-none -fPIC  -O3 -funroll-loops
>>>> -c chksum_tiled.f
>>>> In file chksum_tiled.f:361
>>>> 
>>>> These lines are here to deliberately cause a compile-time error.
>>>> If you see these lines in your .F files or the compiler shows them
>>>> as an error then it means you have not placed your configuration
>>>> files in the appropriate place.
>>>> You need to place you own copy of SIZE.h in the include
>>>> path for the model, and comment out these lines.
>>>> 
>>> 
>>> 
>>> Martin
>>> 
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>> 
>> 
>> 
>> --
>> A.Nazemian
>> 
> 
> 
> -- 
> A.Nazemian
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support