Top Page > Browsing
The memory size can be reduced by a hybrid parallel method
Date: 2018/03/23 11:50
Name: furtrue   <1606036223@qq.com>

Dear OpenMX developers and users,
I want to know that in OPENMX, we use different parallel ways (pure MPI, mpi/openmp) to test the example, and how to compare the occupancy of memory size. Why can we reduce memory size by mixing method?


Please let me know the solution for this.


Thanking you,

メンテ
Page: [1]

Re: The memory size can be reduced by a hybrid parallel method ( No.1 )
Date: 2018/03/23 21:41
Name: T. Ozaki

Hi,

> how to compare the occupancy of memory size

A simple way is to use the built-in tool of OpenMX as explained at
http://www.openmx-square.org/openmx_man3.8/node171.html
However, the tool does not give the precise analysis of memory usage.
So, I recommend you to use tools of performance analysis such as VTune
and CrayPat provided by vender.

> Why can we reduce memory size by mixing method?

As the number of MPI processes increases, buffer memory used for MPI communications
will also be allocated, resulting in increase of the total memory requirement.
Also some of arrays are not properly parallelized for the conventional diagonalization,
which largely increases the memory requirement.

For the latter case, by using the ScaLAPACK version one can reduce the memory requirement.
Take a look at
http://www.openmx-square.org/openmx_man3.8/node9.html
http://www.openmx-square.org/openmx_man3.8/node88.html


Regards,

TO
メンテ

Page: [1]

Thread Title (must) Move the thread to the top
Your Name (must)
E-Mail (must)
URL
Password (used in modification of the submitted text)
Comment (must)

   Save Cookie