Hello/bonjour,
I modified your patch so that in the worse case exactly comm->num_procs are
probed (I just changed a < for a <= in the second for loop).
The behavior of your patch was to probe a maximum of comm->num_procs - 1, thus
hanging if comm->last_probed has to be probed.
With the patch
Sebastien,
Please try the attached patch. It is made against the trunk, so you might have
to adapt it slightly.
Let me know of the outcome, so we can decide if it's worth pushing it in the
next releases.
Thanks,
george.
iprobe_starvation.patch
Description: Binary data
On Sep 27, 201
On Sep 27, 2011, at 16:21 , Sébastien Boisvert wrote:
> Hello George,
>
>
> Thank you for reply.
>
> I have read the OB1 code.
>
>
> Let's suppose that the source is MPI_ANY_SOURCE.
>
> And let's suppose that we have this call stack:
>
>
> 1. MPI_Iprobe
> (in ompi/mpi/c/iprobe.c)
> 2. mca
Hello George,
Thank you for reply.
I have read the OB1 code.
Let's suppose that the source is MPI_ANY_SOURCE.
And let's suppose that we have this call stack:
1. MPI_Iprobe
(in ompi/mpi/c/iprobe.c)
2. mca_pml_ob1_iprobe
(in ompi/mca/pml/ob1/pml_ob1_iprobe.c)
3. mca_pml_ob1_recv_req_start
Sebastien,
Your analysis is correct in case the checkpoint/restart approach maintained by
ORNL is enabled. This is not the code path of the "normal" MPI processes, where
the PML OB1 is used. In this generic case the function mca_pml_ob1_iprobe,
defined in the file ompi/mca/pml/ob1/pml_ob1_iprob
Hello,
As I understand, When MPI_Iprobe is called, the code that is called is the
function pointed by the attribute
mca_pml_base_module_iprobe_fn_t pml_iprobe;
in ompi/mca/pml/pml.h
In the file ompi/mca/crcp/bkmrk/crcp_bkmrk_pml.c (Open-MPI 1.4.3),
ompi_crcp_bkmrk_pml_iprobe calls dr
In the usual place:
http://www.open-mpi.org/software/ompi/v1.4/
Please test!
--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
On Sep 27, 2011, at 5:43 AM, Xin He wrote:
> sorry that it takes so long for paper works of TIPC BTL. Hopefully I can get
> some feedback this week.
Great.
> However, I would like to do some work while waiting. For future work, I would
> like to add TIPC_broadcast
> to support MPI_Bcast. But a
Xin,
The collective calls are handled by the collective framework (mca/coll). You
can export only the function you want (MPI_Bcast in your example) from your
collective component.
george.
On Sep 27, 2011, at 05:43 , Xin He wrote:
> Hi all,
>
> sorry that it takes so long for paper works of
Hi all,
sorry that it takes so long for paper works of TIPC BTL. Hopefully I can
get some feedback this week.
However, I would like to do some work while waiting. For future work, I
would like to add TIPC_broadcast
to support MPI_Bcast. But as I looked into TCP BTL code, I could not
find any
10 matches
Mail list logo
