From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Fri Nov 20 2015 - 08:02:36 CST
The MPI version you have been building should work, and with replicas you
probably won't see much of a performance difference.
Who made your IB cards? They may now be supported directly, either in the
released binaries or with a recompilation.
The requirement for replicas is mpi, verbs, netlrts, or gni, so in your
case you would want (copying from the release notes):
Build and test the Charm++/Converse library (InfiniBand version):
./build charm++ verbs-linux-x86_64 --with-production
./charmrun ++mpiexec +p4 ./pgm (uses mpiexec to launch processes)
./config Linux-x86_64-g++ --charm-arch verbs-linux-x86_64
make (or gmake -j4, which should run faster)
If you have Intel compilers use those (verbs-linux-x86_64-iccstatic and
On Fri, 20 Nov 2015, Maxime Boissonneault wrote:
> A user of our cluster asked for us to install
> His need is to be able to perform replica exchanges, and he says that only
> this version allowed him to do so. We have an older 2.9 version of NAMD that
> we compiled, and apparently it does not support replica exchanges.
> The thing is, we have a cluster with non-blocking QDR Infiniband. We usually
> compile Namd from source, with MPI.
> What version would you recommend me to install, and if it's a compilation
> from source, what should be the compilation options to enable replica
> exchanges ?
> Maxime Boissonneault
> Analyste de calcul - Calcul Québec, Université Laval
> Président - Comité de coordination du soutien à la recherche de Calcul
> Team lead - Research Support National Team, Compute Canada
> Instructeur Software Carpentry
> Ph. D. en physique
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:21:34 CST