Dear Einstein Toolkit users

Up until now, Simfactory did not use OpenMP by default. That is, even if a code was built with OpenMP, Simfactory would by default use only a single thread when submitting a job, and parallelise purely via MPI instead. This reflected the state of the thorns, as several key thorns could in the past not make efficient use of OpenMP.

This has changed, and e.g. the recent release of the Einstein Toolkit contains efficient OpenMP-parallelised implementations of the BSSN and GR-MHD equations, and all other key thorns. To our knowledge, most people use OpenMP for most of their production runs.

We therefore now proudly announce that we are enabling OpenMP in job submissions by default. Simfactory "knows" what a good number of OpenMP threads is for most machines. This number depends on the hardware architecture, and thus differs from machine to machine, and Simfactory will use that many threads by default. You can override this via the --num-threads=... option; using a single thread will effectively disable OpenMP.

I assume that this "good" number of threads described in Simfactory may require some adjustment in some cases. Please report cases where you think that Simfactory is mis-behaving.

-erik