Thank you, Gilles and Jeff. This makes a lot of sense now.
And, Jeff, I thnk the paper you mentioned is this
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5184825&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D5184825
?
Thank you,
Slaiya
On Mon, May 30, 2016 at 9:
> So, you mean that it guarantees the value received after the bcast call is
> consistent with value sent from root, but it doesn't have to wait till all
> the ranks have received it?
>
> this is what i believe, double checking the standard might not hurt though
> ...
>
No function has barrier sem
On 5/30/2016 11:09 PM, Saliya Ekanayake wrote:
So, you mean that it guarantees the value received after the bcast
call is consistent with value sent from root, but it doesn't have to
wait till all the ranks have received it?
this is what i believe, double checking the standard might not hurt
Hi,
in your .bashrc, Open MPI environment (PATH and LD_LIBRARY_PATH) is set
at the very end.
i do not know who invokes mpicc, but it is very likely invoked *before*
the environment is set.
/* note this might not be the mpicc you expect */
i would first double check
~/OpenFOAM/OpenFOAM-v3
Hello Everyone,
I have attempted to upgrade my Open MPI to version 1.10.2 however after
following the build steps on the FAQ page
(https://www.open-mpi.org/faq/?category=building) I receive the following error
at the top of every terminal I open, "mpicc: error while loading shared
libraries: l
So, you mean that it guarantees the value received after the bcast call is
consistent with value sent from root, but it doesn't have to wait till all
the ranks have received it?
Still, in this benchmark shouldn't the max time for bcast be equal to that
of barrier?
On Mon, May 30, 2016 at 9:33 AM,
These are very different algorithms, so performance might differ (greatly)
for example, MPI_Bcast on root rank can MPI_Send() and return, if the
message is short, this is likely an eager send which is very fast.
that means MPI_Bcast() returns before all ranks received the data, or even
entered MPI
These were taken using OSU Micro benchmarks 5.3
http://mvapich.cse.ohio-state.edu/benchmarks/
For example, in a cluster with 32 nodes each running 24 processes,
Broadcast for bytes 1 to 64 take around 36 us where as the barrier takes
165 us. These were on 40Gbps Infiniband
# OSU MPI Broadcast La
Hi,
How are you measuring these times?
Thanks,
Matthieu
From: users [users-boun...@open-mpi.org] on behalf of Saliya Ekanayake
[esal...@gmail.com]
Sent: Monday, May 30, 2016 7:53 AM
To: Open MPI Users
Subject: [OMPI users] Broadcast faster than barrier
Hi,
I
Hi,
I ran Ohio micro benchmarks for openmpi and noticed broadcast with smaller
number of bytes is faster than a barrier - 2us vs 120us.
I'm trying to understand how this could happen?
Thank you
Saliya
10 matches
Mail list logo
