REMD with distance restrance
Created an attachment (id=567)
I am trying to do REMD with distance restraints using gromacs-4.5.1. My problem is that if i specify the number of cores which are more than the number of tpr files that i have, my simulations are crushing with a segmentation fault. They seem to run very well if i specify the number of processors which are equal to the number of tpr files . I believe that is how REMD was meant work in versions prior to gromacs-4.0. If i run the same simulation without distance restraints, i can specify as many cores as i want and the simulations seem to run very well.
I am wondering if there is anything i am doing wrong,
I have tried to show the commands that i am using below...i am working with 16 replicas
1. mpprun -np 16 -pd -s md_tpr -multi 16 -replex 5000 -o md_
This one works fine with distance restraints
2. mpprun -np 64 -pd -s md_tpr -multi 16 -replex 5000 -o md_
The second one works fine without distance restraints but crushes with a segmentation fault when i use distance restraints.
My system is made up of two similar peptide molecules in a box. I have 36 nmr distance restraints on each of the peptides.
Thank you for your help
Updated code checks with distance restraints
Ensemble-averaged distance restraints require -multi, and it is
also reasonable to do REMD with distance restraints, which also
requires -multi. So checks for the ensemble-averaging case need
to be more sensitive to their context.
In fact, ensemble-averaging is probably functional only with PD and
one processor per system, since that was probably all that was
available when it was built.
#3 Updated by David van der Spoel almost 10 years ago
I fixed a simple bug in the printing which caused the SEGV (commit 74c3d74 in the release branch), however it seems that parallel DR is not compatible with REMD yet, since there is another crash, now in the MPI routine Allreduce. Will check more.