Hello,
I'm working on a collective component and need point-to-point
communication during module-initialization.
As BTL is initialized prior to the collectives, I tried to use send and
recv like MPI_Send/_Recv do:
err = MCA_PML_CALL(send(buf, size, MPI_CHAR, to_id,
COLL_SCI_TAG, MCA_P
Hello,
When running the Intel MPI Benchmark (IMB) on our cluster
(Sun X2200M2 nodes, Voltaire DDRx Infiniband, OFED-1.1) we
see rather strange (i.e., unreasonably bad) performance for the
Allgatherv part of the IMP when using OpenMPI-1.2.2. The
performance figures reported by the IMB are provided
I assume you mean something like mca_coll_foo_init_query() for your
initialization function. And I'm guessing you're exchanging some sort
of address information for your network here?
What I actually did in my collective component was use PML's modex
(module exchange) facility, defined in
om
Hello all,
I temporarily worked around my former problem by using synchronous
communication and shifting the initialization
into the first call of a collective operation.
But nevertheless, I found a performance bug in btl_openib.
When I execute the attached sendrecv.c on 4 (or more) nodes of
On Fri, Jun 22, 2007 at 04:52:45PM -0400, Jeff Squyres wrote:
> On Jun 20, 2007, at 8:29 AM, Jeff Squyres wrote:
>
> >1. btl_*_min_send_size is used to decide when to stop striping a
> >message across multiple BTL's. Is there a reason that we don't
> >just use eager_limit for this value? It
Hello... I'm a student at the University of British Columbia working on
creating an SCTP BTL for Open MPI. I have a simple implementation
working that uses SCTPs one-to-one style sockets for sending messages.
The same writev()/readv() calls that are used in the TCP BTL are used in
this new BTL.
Hello Peter,
in 1.2.2 the allgatherv is called from coll basic component,
and is implemented as a gatherv followed by a broadcast.
The broadcast is executed with single element of MPI_TYPE_INDEXED.
The decision function in coll tuned makes mistake of using segmented
broadcast algorithm for th
Karol,
We (the folks at UTK) implemented a SCTP BTL. It's not yet in the
trunk, but it will get there shortly. Instead of starting from
scratch, it might be a good idea to start directly from there.
To answer your question, the TCP BTL use a copy of the original
iovec. After each write, t
Gleb,
Simplifying the code and getting better performance is always a good
approach (at least from my perspective). However, your patch still
dispatch the messages over the BTLs in a round robin fashion, which
doesn't look to me as the best approach. How about merging your patch
and mine
George Bosilca wrote:
Karol,
We (the folks at UTK) implemented a SCTP BTL. It's not yet in the
trunk, but it will get there shortly. Instead of starting from
scratch, it might be a good idea to start directly from there.
Thanks for the reply. This BTL would definitely be worth taking a look
a
On Jun 26, 2007, at 6:08 PM, Tim Prins wrote:
Some time ago you were working on moving the modex out of the pml
and cleaning
it up a bit. Is this work still ongoing? The reason I ask is that I am
currently working on integrating the RSL, and would rather build on
the new
code rather than th
11 matches
Mail list logo
