[mpich-discuss] discuss Digest, Vol 9, Issue 10
Sufeng Niu
sniu at hawk.iit.edu
Fri Jul 5 11:52:48 CDT 2013
Thanks a lot! Jim. Now I make sense
Sufeng
On Fri, Jul 5, 2013 at 11:25 AM, <discuss-request at mpich.org> wrote:
> Send discuss mailing list submissions to
> discuss at mpich.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://lists.mpich.org/mailman/listinfo/discuss
> or, via email, send a message with subject or body 'help' to
> discuss-request at mpich.org
>
> You can reach the person managing the list at
> discuss-owner at mpich.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of discuss digest..."
>
>
> Today's Topics:
>
> 1. Re: beginner for remote mem access (sufeng) (Pavan Balaji)
> 2. MPI Benchmark Suite (Jiri Simsa)
> 3. Re: beginner for remote mem access (sufeng) (Jim Dinan)
> 4. Re: beginner for remote mem access (sufeng) (Sufeng Niu)
> 5. Re: beginner for remote mem access (sufeng) (Pavan Balaji)
> 6. Re: MPI Benchmark Suite (Pavan Balaji)
> 7. Re: beginner for remote mem access (sufeng) (Sufeng Niu)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Fri, 05 Jul 2013 10:36:07 -0500
> From: Pavan Balaji <balaji at mcs.anl.gov>
> To: discuss at mpich.org
> Cc: Sufeng Niu <sniu at hawk.iit.edu>
> Subject: Re: [mpich-discuss] beginner for remote mem access (sufeng)
> Message-ID: <51D6E7E7.8070306 at mcs.anl.gov>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
>
> On 07/05/2013 10:28 AM, Sufeng Niu wrote:
> > 1.
> > MPI_Win_fence is not similar to MPI_Barrier in some cases. The
> > MPICH implementation can turn some calls into no-ops and others into
> > MPI_Reduce_scatter.
> > I am not quite sure about "turn some calls into no-ops and others into
> > MPI_Reduce_scatter" Could you please give an example if possible?
> > Another thing is if a process create a window for other process, but the
> > data access is available after some operations. Should I use
> > MPI_Win_fence or MPI_Barrier to sync? or other methods?
>
> MPI_WIN_FENCE will make sure that all PUT/GET/ACCUMULATE operations are
> complete locally and remotely. This is probably what you need.
>
> MPI_BARRIER is useless in this case since PUT/GET/ACCUMULATE can be
> nonblocking.
>
> > 2.
> > Regarding "use MPI window as thread level", I really don't
> > understand your thinking at all. MPI RMA is not shared memory nor is
> > MPI a threading model.
> > Sorry for uncleared statements, right now I would like to do
> > multithreads and MPI hybrid programming. let me give an example: I have
> > 3 processes, each one has 8 threads. thread 0 in process 0 creates a RMA
> > window. if I would like all other threads to access it, should I use
> > thread 0 in process 1, thread 0 in process 2 to MPI_Get the data from
> > window, then use shared memory for internal threads to load the data? i
> > am not sure what is the proper way for RMA in hybrid model.
>
> All threads within the OS process are part of the same MPI rank. Any of
> them can access the window as the same rank.
>
> -- Pavan
>
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji
>
>
> ------------------------------
>
> Message: 2
> Date: Fri, 5 Jul 2013 11:38:13 -0400
> From: Jiri Simsa <jsimsa at cs.cmu.edu>
> To: discuss at mpich.org
> Subject: [mpich-discuss] MPI Benchmark Suite
> Message-ID:
> <CAHs9ut-=
> FJeUDG-T2-qqDhcDMBGfGuXPHG7VQp8_xYzXTxPAzA at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hello,
>
> Could anyone point me in the direction of programs that are representative
> of MPI programs? I am looking for something like PARSEC (
> http://parsec.cs.princeton.edu/) for MPI. In other words, I am interested
> in macro-bechnmarks not micro-benchmarks. Thank you.
>
> Best,
>
> --Jiri Simsa
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> http://lists.mpich.org/pipermail/discuss/attachments/20130705/ab3a6a55/attachment-0001.html
> >
>
> ------------------------------
>
> Message: 3
> Date: Fri, 5 Jul 2013 11:43:29 -0400
> From: Jim Dinan <james.dinan at gmail.com>
> To: discuss at mpich.org
> Subject: Re: [mpich-discuss] beginner for remote mem access (sufeng)
> Message-ID:
> <
> CAOoEU4E1j4x7hxR6FrZMLrjsfaH31tzDTgucyiF5F3osduxY2Q at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> On Fri, Jul 5, 2013 at 11:28 AM, Sufeng Niu <sniu at hawk.iit.edu> wrote:
>
> > 1.
> > MPI_Win_fence is not similar to MPI_Barrier in some cases. The
> > MPICH implementation can turn some calls into no-ops and others into
> > MPI_Reduce_scatter.
> > I am not quite sure about "turn some calls into no-ops and others into
> > MPI_Reduce_scatter" Could you please give an example if possible? Another
> > thing is if a process create a window for other process, but the data
> > access is available after some operations. Should I use MPI_Win_fence or
> > MPI_Barrier to sync? or other methods?
> >
>
> When you call MPI_Win_fence, the MPI implementation does not need to wait
> for all processes to call the function before proceeding. It does require
> that all RMA operations in a given process' access epoch are complete, and
> that all RMA accesses in its exposure epoch (i.e., targeting the given
> process) are also complete. If, for example, there is no preceding epoch,
> the call to MPI_Win_fence is a no op.
>
> Generally speaking, you should synchronize all accesses to data exposed in
> a window using RMA synchronization operations. Barrier does not provide
> any consistency for RMA operations.
>
>
> > 2.
> > Regarding "use MPI window as thread level", I really don't
> > understand your thinking at all. MPI RMA is not shared memory nor is
> > MPI a threading model.
> > Sorry for uncleared statements, right now I would like to do multithreads
> > and MPI hybrid programming. let me give an example: I have 3 processes,
> > each one has 8 threads. thread 0 in process 0 creates a RMA window. if I
> > would like all other threads to access it, should I use thread 0 in
> process
> > 1, thread 0 in process 2 to MPI_Get the data from window, then use shared
> > memory for internal threads to load the data? i am not sure what is the
> > proper way for RMA in hybrid model.
> >
>
> The threads can all access the data in the window, provided that the MPI
> process synchronizes accesses by threads in the process correctly with
> respect to what other processes are doing. Note that data consistency
> across threads is outside of the MPI specification, and you should use e.g.
> Pthreads or OpenMP primitives to ensure that data is consistent across
> threads.
>
> ~Jim.
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> http://lists.mpich.org/pipermail/discuss/attachments/20130705/d2064a55/attachment-0001.html
> >
>
> ------------------------------
>
> Message: 4
> Date: Fri, 5 Jul 2013 11:18:22 -0500
> From: Sufeng Niu <sniu at hawk.iit.edu>
> To: Pavan Balaji <balaji at mcs.anl.gov>
> Cc: discuss at mpich.org
> Subject: Re: [mpich-discuss] beginner for remote mem access (sufeng)
> Message-ID:
> <
> CAFNNHkyV3k1168kYjOhX2OpeF+TcjLRsg6Oj2oL0pZ-KwHQKXw at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi Pavan,
>
> Thanks a lot!
>
> Now I make sense for MPI_Win_fence.
>
> for hybrid RMA issue, did you mean in each rank, remote get the data into a
> shared memory, then threads in each rank load it by thread-level operation?
>
> Thank you!
> Sufeng
>
>
>
> On Fri, Jul 5, 2013 at 10:36 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>
> >
> > On 07/05/2013 10:28 AM, Sufeng Niu wrote:
> >
> >> 1.
> >> MPI_Win_fence is not similar to MPI_Barrier in some cases. The
> >> MPICH implementation can turn some calls into no-ops and others into
> >> MPI_Reduce_scatter.
> >> I am not quite sure about "turn some calls into no-ops and others into
> >> MPI_Reduce_scatter" Could you please give an example if possible?
> >> Another thing is if a process create a window for other process, but the
> >> data access is available after some operations. Should I use
> >> MPI_Win_fence or MPI_Barrier to sync? or other methods?
> >>
> >
> > MPI_WIN_FENCE will make sure that all PUT/GET/ACCUMULATE operations are
> > complete locally and remotely. This is probably what you need.
> >
> > MPI_BARRIER is useless in this case since PUT/GET/ACCUMULATE can be
> > nonblocking.
> >
> >
> > 2.
> >> Regarding "use MPI window as thread level", I really don't
> >> understand your thinking at all. MPI RMA is not shared memory nor is
> >> MPI a threading model.
> >> Sorry for uncleared statements, right now I would like to do
> >> multithreads and MPI hybrid programming. let me give an example: I have
> >> 3 processes, each one has 8 threads. thread 0 in process 0 creates a RMA
> >> window. if I would like all other threads to access it, should I use
> >> thread 0 in process 1, thread 0 in process 2 to MPI_Get the data from
> >> window, then use shared memory for internal threads to load the data? i
> >> am not sure what is the proper way for RMA in hybrid model.
> >>
> >
> > All threads within the OS process are part of the same MPI rank. Any of
> > them can access the window as the same rank.
> >
> > -- Pavan
> >
> > --
> > Pavan Balaji
> > http://www.mcs.anl.gov/~balaji
> >
>
>
>
> --
> Best Regards,
> Sufeng Niu
> ECASP lab, ECE department, Illinois Institute of Technology
> Tel: 312-731-7219
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> http://lists.mpich.org/pipermail/discuss/attachments/20130705/2454c8c0/attachment-0001.html
> >
>
> ------------------------------
>
> Message: 5
> Date: Fri, 05 Jul 2013 11:20:41 -0500
> From: Pavan Balaji <balaji at mcs.anl.gov>
> To: Sufeng Niu <sniu at hawk.iit.edu>
> Cc: discuss at mpich.org
> Subject: Re: [mpich-discuss] beginner for remote mem access (sufeng)
> Message-ID: <51D6F259.80800 at mcs.anl.gov>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
>
> On 07/05/2013 11:18 AM, Sufeng Niu wrote:
> > for hybrid RMA issue, did you mean in each rank, remote get the data
> > into a shared memory, then threads in each rank load it by thread-level
> > operation?
>
> All threads are in the same process, so they all already share memory.
> Perhaps its best if you try to learn how threads work before attempting
> to use them with MPI.
>
> -- Pavan
>
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji
>
>
> ------------------------------
>
> Message: 6
> Date: Fri, 05 Jul 2013 11:22:28 -0500
> From: Pavan Balaji <balaji at mcs.anl.gov>
> To: discuss at mpich.org
> Cc: Jiri Simsa <jsimsa at cs.cmu.edu>
> Subject: Re: [mpich-discuss] MPI Benchmark Suite
> Message-ID: <51D6F2C4.7030002 at mcs.anl.gov>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
>
> On 07/05/2013 10:38 AM, Jiri Simsa wrote:
> > Could anyone point me in the direction of programs that are
> > representative of MPI programs? I am looking for something like PARSEC
> > (http://parsec.cs.princeton.edu/) for MPI. In other words, I am
> > interested in macro-bechnmarks not micro-benchmarks. Thank you.
>
> You could try the Sequoia suite. There's also the NAS parallel
> benchmarks and Graph500 benchmarks.
>
> -- Pavan
>
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji
>
>
> ------------------------------
>
> Message: 7
> Date: Fri, 5 Jul 2013 11:24:59 -0500
> From: Sufeng Niu <sniu at hawk.iit.edu>
> To: Pavan Balaji <balaji at mcs.anl.gov>
> Cc: discuss at mpich.org
> Subject: Re: [mpich-discuss] beginner for remote mem access (sufeng)
> Message-ID:
> <
> CAFNNHkzCaNw+gQhSriR1xUOo853bNPAtM7Z3TTOBS_XB1JD3ww at mail.gmail.com>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Thanks a lot!
>
> Now I got it.
>
> Sufeng
>
>
> On Fri, Jul 5, 2013 at 11:20 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>
> >
> > On 07/05/2013 11:18 AM, Sufeng Niu wrote:
> >
> >> for hybrid RMA issue, did you mean in each rank, remote get the data
> >> into a shared memory, then threads in each rank load it by thread-level
> >> operation?
> >>
> >
> > All threads are in the same process, so they all already share memory.
> > Perhaps its best if you try to learn how threads work before attempting
> to
> > use them with MPI.
> >
> >
> > -- Pavan
> >
> > --
> > Pavan Balaji
> > http://www.mcs.anl.gov/~balaji
> >
>
>
>
> --
> Best Regards,
> Sufeng Niu
> ECASP lab, ECE department, Illinois Institute of Technology
> Tel: 312-731-7219
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> http://lists.mpich.org/pipermail/discuss/attachments/20130705/f0eefadd/attachment.html
> >
>
> ------------------------------
>
> _______________________________________________
> discuss mailing list
> discuss at mpich.org
> https://lists.mpich.org/mailman/listinfo/discuss
>
> End of discuss Digest, Vol 9, Issue 10
> **************************************
>
--
Best Regards,
Sufeng Niu
ECASP lab, ECE department, Illinois Institute of Technology
Tel: 312-731-7219
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpich.org/pipermail/discuss/attachments/20130705/a674adb4/attachment.html>
More information about the discuss
mailing list