Charm++ parallel programming system

[David Fallaize <d.fallaize AT ucl.ac.uk>]

 Hi,

I've written my parallel program in charm++ which is intended to run on 
a typical sort of cluster where each node is a multi-core machine, 
there's a job submission queue etc. etc. which basically means I build 
mpi-linux-x86_64-smp-mpicxx, using MPICH2.  I'm using the latest version 
of Charm from Git.

From the description of nodegroups, I expect one of these objects to be 
instantiated on each node I run on.  However when I run my program I 
notice that I get one nodegroup object created per processor, not per 
node.  I checked using one of the test programs which comes with charm 
(pingpong) and by adding a line to the PingN constructor (which is 
supposed to be a nodegroup object) I can observe 2 of these being 
created on a single SMP node: see "Creating a PingN" appears twice in 
the output appended below.  (I tried using OpenMPI too, and got the same 
result).

Using the net-smp version (which is what I've been doing on my 
development machine) gives a single nodegroup per node as expected, so 
why do I not get the same under MPI?  Is this a bug?  This is a serious 
problem for me as part of my program rather relies on the one-per-node 
behaviour of nodegroups!

Thanks,

David Fallaize
PhD student, UCL, London

Running on 2 processors:  ./pgm
charmrun>  /usr/bin/setarch x86_64 -R  mpirun -np 2  ./pgm
Charm++> Running on MPI version: 2.1 multi-thread support: 
MPI_THREAD_FUNNELED (max supported: MPI_THREAD_SINGLE)
Converse/Charm++ Commit ID: v6.3.0-308-gbb91f75
Charm++> Running on 1 unique compute nodes (8-way SMP).
Creating a PingN.
Charm++> cpu topology info is gathered in 0.000 seconds.
Pingpong with payload: 100 iterations: 1000
Creating a PingN.
Roundtrip time for 1D Arrays is 26.417017 us
Roundtrip time for 1D threaded Arrays is 30.620098 us
Roundtrip time for 2D Arrays is 26.331186 us
Roundtrip time for 3D Arrays is 26.363850 us
Roundtrip time for Fancy Arrays is 26.504993 us
Roundtrip time for Chares (reuse msgs) is 25.969982 us
Roundtrip time for Chares (new/del msgs) is 26.633024 us
Roundtrip time for threaded Chares (reuse) is 30.007124 us
Roundtrip time for Groups is 26.214123 us
Roundtrip time for NodeGroups is 26.480913 us
End of program