How To Compile and runn Elk on a multi-node server

In this HowTo we describe how Elk should be compiled for maximum performance on a multi-node cluster.

Elk uses both the Message Passing Interface (MPI) and OpenMP for parallelism as well as highly optimised libraries for linear algebra and the fast Fourier transform (FFT). Compilation may also include linking to the Libxc and Wannier90 libraries.

We will also show how to run the code across multiple nodes using the mpirun command and which OpenMP environment variables should be set.

Considerable effort is put into optimising the Elk and each new version is usually faster than the last. Always try to compile your own code from scratch using latest official release (found here http://sourceforge.net/projects/elk/files/) rather than using a pre-compiled package. This way the executable will be tailored for your particular computer system.

For this HowTo we will use Elk version 7.1.14. Download elk-7.1.14.tgz and unpack it.

 

https://sourceforge.net/p/elk/discussion/897820/thread/bb25c53e70/