Boost logo

Boost-Build :

Subject: Re: [Boost-build] Patch to serialize MPI tests
From: Belcourt, Kenneth (kbelco_at_[hidden])
Date: 2010-05-31 11:13:57


On May 29, 2010, at 8:47 PM, Jeremiah Willcock wrote:

> On Fri, 14 May 2010, Matthias Troyer wrote:
>
>>
>> On 14 May 2010, at 20:04, Belcourt, Kenneth wrote:
>>> The attached patch serializes the execution of the MPI tests. I
>>> need to serialize the execution of the MPI tests so that when
>>> running Boost regression tests in parallel, we avoid overloading
>>> our systems.
>>>
>>> [kbelco_at_wsblade001 tools]$ svn diff
>>> Index: mpi.jam
>>> ===================================================================
>>> --- mpi.jam (revision 61969)
>>> +++ mpi.jam (working copy)
>>> @@ -493,6 +493,9 @@
>>> local num_processes = [ property.select <mpi:processes> : $
>>> (properties) ] ;
>>> num_processes = $(num_processes:G=) ;
>>>
>>> + # serialize the MPI tests to avoid overloading systems
>>> + JAM_SEMAPHORE on $(target) = <s>mpi-run-semaphore ;
>>> +
>>> # We launch MPI processes using the "mpirun" equivalent
>>> specified by the user.
>>> LAUNCHER on $(target) =
>>> [ on $(target) return $(.mpirun) $(.mpirun_flags) $
>>> (num_processes) ] ;
>>>
>>>
>>> This has been tested on two systems (Darwin and Linux) without any
>>> problems. Okay to commit?
>>
>> As far as I am concerned you can commit this.
>
> Does this patch also serialize the Boost.Graph.Parallel tests? If it
> doesn't, is there a way to have those serialized as well? That may
> avoid
> some of the timeouts that often cause the Boost.Graph.Parallel
> regression
> tests to fail.

Yes, this patch will serialize any tests that use MPI including
graph_parallel.

I did notice an internal compiler error with gcc-4.5 with graph-
parallel. Has that problem been reported to gcc?

http://tinyurl.com/2whzwkv

-- Noel


Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk