Re: Discussion: Switching Esprseso to shared memory parallelization

[Rudolf Weeber]

Dear colleagues,

thank you for your feedback with regards to the discussion about Espresso's 
parallelization.

After further debate, we decided to keep the MPI parallelization so as to keep 
the door open for the simulation of large systems.
Having said that, the majority of the simulations currently performed with 
Espresso run on less than 10 cores. Optimizing for that scale is therefore 
where we will put most of our effort.

When new features are added to Espresso, we will strive to offer a distributed 
memory parallelized implementation. In those cases, where the effort for 
parallelization is disproportionate, we may, however, defer the 
parallelization, until a specific use cases comes up.

We are currently investigating using performance portability libraries to gain 
more flexibility. This is not a short-term endeavour, though.

Regards, Rudolf

-- 
Dr. Rudolf Weeber
Institute for Computational Physics
Universität Stuttgart
Allmandring 3
70569 Stuttgart
Germany
Phone: +49(0)711/685-67717
Email: weeber@icp.uni-stuttgart.de
http://www.icp.uni-stuttgart.de/~icp/Rudolf_Weeber