Re: [Beowulf] mpi_comm_create scaling issue

2013-09-05 Thread Christopher Samuel
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 06/09/13 01:56, xingqiu yuan wrote: > MPI_COMM_CREATE takes a substantial amount of time on large > communicators, any good ideas to reduce the time consuming on > large communicators? Which version of which MPI stack is this, or is this across m

Re: [Beowulf] mpi_comm_create scaling issue

2013-09-05 Thread Lawrence Stewart
Are you doing this dynamically? Could you change the code to create the necessary communicators just once at job launch time? -L On 2013, Sep 5, at 11:56 AM, xingqiu yuan wrote: > Hi ALL > > MPI_COMM_CREATE takes a substantial amount of time on large communicators, > any good ideas to reduce

[Beowulf] mpi_comm_create scaling issue

2013-09-05 Thread xingqiu yuan
Hi ALL MPI_COMM_CREATE takes a substantial amount of time on large communicators, any good ideas to reduce the time consuming on large communicators? regards Xingqiu Yuan ___ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To c

Re: [Beowulf] mpi_comm_create scaling issue

2013-09-05 Thread Charlie Peck
On Sep 5, 2013, at 11:56 AM, xingqiu yuan wrote: > Hi ALL > > MPI_COMM_CREATE takes a substantial amount of time on large communicators, > any good ideas to reduce the time consuming on large communicators? Which MPI binding you use can make a difference too, try another one (or two) and see