[FLASH-USERS] Weak scaling vs Strong scaling of FLASH benchmark?

[Fei Meng]

Hi Anshu,

Thanks for your reply.

I read the source code and found the "io_writeData.F90" controls the
checkpoint file to be a fixed size(for Sod default configuration). I found
every process/node is sending data back to the head node, but I don't know
how the head node could control the checkpoint file size to be constant.
Also when I run over 400 nodes in Sod default configuration, it returns
"MPI_Barrier" error to me.

I am very interested in how to configure the benchmark to be weak scaling. I
would appreciate it very much if you could give me more insight of the
architecture/data flow of the benchmark.

Thansk again.

Fei Meng
Department of Computer Science
North Carolina University

On Sun, Apr 4, 2010 at 3:23 PM, Anshu Dubey <dubey at flash.uchicago.edu>wrote:

> Dear Fei,
>
> If you only change the number of MPI processes, the code operates in strong
> scaling mode. If you wish to do weak scaling studies you have to change the
> problem size by hand. We can give you ideas on how to do weak scaling if you
> are interested.
>
> Anshu
>
> On Thu, Apr 1, 2010 at 11:23 PM, Fei Meng <fmeng at ncsu.edu> wrote:
>
>> Hi there,
>>
>> I am quite new to the FLASH code and I have a very basic question about
>> problem size. Is the FLASH code weak scaling or strong scaling? Putting it
>> in the another way, is the problem size(computation etc.) fixed when I
>> increased the MPI processes?
>>
>> I am doing one test using FLASH but I am very new to Fortran code and the
>> benchmark. Thanks very much. Looking forward to your reply.
>>
>> Fei Meng
>> Department of Computer Science
>> North Carolina University
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20100404/5af519fe/attachment.htm>