[mpich-discuss] discuss Digest, Vol 8, Issue 39

Antonio J. Peña apenya at mcs.anl.gov
Mon Jun 24 16:36:07 CDT 2013 

Thanks Jeff, it looks trivial now you explained it.


On Monday, June 24, 2013 04:30:57 PM Jeff Hammond wrote:
> N threads calling MPI_Barrier corresponds to N different, unrelated
> barriers.  A thread calling MPI_Barrier will only synchronize with
> other processes, not any other threads.
> 
> MPI_Barrier only acts between processes.  It has no effect on threads.
>  Just use comm=MPI_COMM_SELF and think about the behavior of
> MPI_Barrier.  That is the one-process limit of the multithreaded
> problem.
> 
> Jeff
> 
> On Mon, Jun 24, 2013 at 4:22 PM, Antonio J. Peña <apenya at mcs.anl.gov> wrote:
> > Hi Jeff,
> > 
> > Please correct me where I'm wrong: if only one thread calls MPI_Barrier,
> > that one will get locked while others will go ahead. On the other hand,
> > if every thread calls MPI_Barrier, all will get locked in that call until
> > the barrier completes. That doesn't mean this will be a replacement for
> > thread-wise synchronization mechanisms, but will be effective for all of
> > them waiting for the barrier.
> > 
> >   Antonio
> > 
> > On Monday, June 24, 2013 04:14:20 PM Jeff Hammond wrote:
> >> MPI_Barrier will not sync threads.  If N threads call MPI_Barrier, you
> >> will
> >> get at best the same result as if you call MPI_Barrier N times from the
> >> main thread.
> >> 
> >> If you want to sync threads, you need to sync them with the appropriate
> >> thread API.  OpenMP and Pthreads both have barrier calls.  If you want a
> >> fast Pthread barrier, you should not use pthread_barrier though.  The
> >> Internet has details.
> >> 
> >> Jeff
> >> 
> >> ----- Original Message -----
> >> From: "Antonio J. Peña" <apenya at mcs.anl.gov>
> >> To: discuss at mpich.org
> >> Sent: Monday, June 24, 2013 3:49:08 PM
> >> Subject: Re: [mpich-discuss] discuss Digest, Vol 8, Issue 39
> >> 
> >> 
> >> 
> >> 
> >> 
> >> 
> >> Sufeng,
> >> 
> >> 
> >> 
> >> I'd say you're OK syncing your threads by all of them calling
> >> MPI_Barrier,
> >> as it's a thread-safe function.
> >> 
> >> 
> >> 
> >> Antonio
> >> 
> >> 
> >> 
> >> 
> >> 
> >> On Monday, June 24, 2013 11:41:44 AM Sufeng Niu wrote:
> >> 
> >> 
> >> Hi, Antonio
> >> 
> >> 
> >> 
> >> Thanks a lot! Now I make sense. Let's say if I am running MPI and
> >> multithreads program. If I called MPI_Barrier in each threads
> >> 
> >> 
> >> what gonna happen? Will threads be synced by MPI_Barrier? or I should use
> >> thread level sync?
> >> 
> >> 
> >> 
> >> Thank you!
> >> 
> >> 
> >> Sufeng
> >> 
> >> 
> >> 
> >> 
> >> 
> >> 
> >> On Sun, Jun 23, 2013 at 6:54 PM, < 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: Error with MPI_Spawn (Jeff Hammond)
> >> 2. Re: discuss Digest, Vol 8, Issue 37 (Antonio J. Pe?a)
> >> 
> >> 
> >> ----------------------------------------------------------------------
> >> 
> >> Message: 1
> >> Date: Sun, 23 Jun 2013 17:03:09 -0500
> >> From: Jeff Hammond < jeff.science at gmail.com >
> >> To: " discuss at mpich.org " < discuss at mpich.org >
> >> Subject: Re: [mpich-discuss] Error with MPI_Spawn
> >> Message-ID: <-1578386667793986954 at unknownmsgid>
> >> Content-Type: text/plain; charset=ISO-8859-1
> >> 
> >> This is the wrong way to use PETSc and to parallelize a code with a
> >> parallel library in general.
> >> 
> >> Write the PETSc user list and they will explain to you how to
> >> parallelize your code properly with PETSc.
> >> 
> >> Jeff
> >> 
> >> Sent from my iPhone
> >> 
> >> On Jun 23, 2013, at 4:59 PM, Nitel Muhtaroglu < muhtaroglu.n at gmail.com >
> > 
> > wrote:
> >> > Hello,
> >> > 
> >> > I am trying to integrate PETSc library to a serial program. The idea is
> >> > that the serial program creates a linear equation system and then calls
> >> > PETSc solver by MPI_Spawn and then solves this system in parallel. But
> >> > when I execute MPI_Spawn the following error message occurs and the
> >> > solver is not called. I couldn't find a solution to this error. Does
> >> > anyone have an idea about it?
> >> > 
> >> > Kind Regards,
> >> 
> >> ------------------------------
> >> 
> >> Message: 2
> >> Date: Sun, 23 Jun 2013 18:54:38 -0500
> >> From: Antonio J. Pe?a < apenya at mcs.anl.gov >
> >> To: discuss at mpich.org
> >> Subject: Re: [mpich-discuss] discuss Digest, Vol 8, Issue 37
> >> Message-ID: <8872211.IGCl89YM8r at localhost.localdomain>
> >> Content-Type: text/plain; charset="iso-8859-1"
> >> 
> >> 
> >> Sufeng,
> >> 
> >> The correct way is to use the MPI_Init_thread function with
> >> MPI_THREAD_MULTIPLE. This will tell the MPI implementation to be thread
> >> safe. It supports OpenMP and Posix Threads (OpenMP primitives in most
> >> systems are likely to be implemented on top of PThreads).
> >> 
> >> Antonio
> >> 
> >> 
> >> On Sunday, June 23, 2013 11:13:31 AM Sufeng Niu wrote:
> >> 
> >> 
> >> Hi, Antonio
> >> 
> >> 
> >> Thanks a lot for your reply, I just figure out that is the firewall
> >> issue.
> >> after I set the firewall. it works now. Thanks again.
> >> 
> >> 
> >> But I still got a few questions on MPI and multithreads mixed
> >> programming.
> >> Currently, I try to run each process on each server, and each process
> >> using thread pool to run multiple threads (pthread lib). I am not sure
> >> whether it is the correct way or not. I wrote it as:
> >> 
> >> 
> >> MPI_Init()
> >> ....
> >> ...
> >> /* create thread pool and initial */
> >> ......
> >> /* fetch job into thread pool */
> >> ......
> >> 
> >> 
> >> MPI_Finalize();
> >> 
> >> 
> >> When I check the book and notes, I found people use
> >> 
> >> 
> >> MPI_Init_thread() with MPI_THREAD_MULTIPLE
> >> 
> >> 
> >> but the some docs said it supported OpenMP, is that possible to use it
> >> with pthread library?
> >> I am new guy to this hybrid programming. I am not sure which is the
> >> proper
> >> way to do it. Any suggestions are appreciate. Thank you!
> >> 
> >> 
> >> Sufeng
> >> 
> >> 
> >> 
> >> 
> >> On Sat, Jun 22, 2013 at 12:12 PM, < discuss-request at mpich.org [1]> wrote:
> >> 
> >> 
> >> Send discuss mailing list submissions to discuss at mpich.org [2]
> >> https://lists.mpich.org/mailman/listinfo/discuss[3]
> >> discuss-request at mpich.org [1]
> >> discuss-owner at mpich.org [4]
> >> sniu at hawk.iit.edu [5]>To: discuss at mpich.org [2]
> >> CAFNNHkwpqdGfZXctL0Uz3hpeL25mZZMtB93qGXjc_+tjnV4csA at mail.gmail.c
> >> om[6]>Content-Type: text/plain; charset="iso-8859-1"
> >> 
> >> Hi,
> >> 
> >> Sorry to bother you guys on this stupid question. last time I re-install
> >> OSfor all blades to keep them the same version. after I mount, set
> >> keylessssh, the terimnal gives the error below:
> >> 
> >> [ proxy:0:1 at iocfccd3.aps.anl.gov [7]]
> >> HYDU_sock_connect(./utils/sock/sock.c:174): unable to connect from
> >> " iocfccd3.aps.anl.gov [8]" to" iocfccd1.aps.anl.gov [9]" (No route to
> >> host) [ proxy:0:1 at iocfccd3.aps.anl.gov [7]] main
> >> (./pm/pmiserv/pmip.c:189): unable toconnect to server
> >> iocfccd1.aps.anl.gov [9] at port 38242 (check for firewalls!)
> >> 
> >> I can ssh from iocfccd1 to iocfccd3 without password. Should I shut
> >> downall
> >> firewalls on each server? I cannot find out where is the problem.
> >> Thankyou
> >> 
> >> --Best Regards,Sufeng NiuECASP lab, ECE department, Illinois Institute of
> >> TechnologyTel: 312-731-7219[10]
> >> http://lists.mpich.org/pipermail/discuss/attachments/20130621/5503b1bc/a
> >> ttachment-0001.html[11] >
> >> 
> >> ------------------------------
> >> 
> >> Message: 2Date: Fri, 21 Jun 2013 10:58:26 -0500From: Antonio J. Pe?a
> >> < apenya at mcs.anl.gov [12]>To: discuss at mpich.org [2]
> >> iocfccd1.aps.anl.gov [9]" from iocfccd3?[1]
> >> 
> >> Antonio
> >> 
> >> 
> >> On Friday, June 21, 2013 10:51:50 AM Sufeng Niu wrote:
> >> 
> >> 
> >> Hi,
> >> 
> >> 
> >> Sorry to bother you guys on this stupid question. last time I re-install
> >> OS
> >> forall blades to keep them the same version. after I mount, set keyless
> >> ssh,the terimnal gives the error below:
> >> 
> >> 
> >> proxy:0:1 at iocfccd3.aps.anl.gov [7][2]]
> >> HYDU_sock_connect(./utils/sock/sock.c:174): unable to connect from
> >> " iocfccd3.aps.anl.gov [8][3]"to " iocfccd1.aps.anl.gov [9][1]" (No route
> >> to host)[ proxy:0:1 at iocfccd3.aps.anl.gov [7][2]] main
> >> (./pm/pmiserv/pmip.c:189):unable to connect to server
> >> iocfccd1.aps.anl.gov [9][1] at port 38242 (checkfor firewalls!)
> >> 
> >> 
> >> 
> >> I can ssh from iocfccd1 to iocfccd3 without password. Should I shut
> >> downall
> >> firewalls on each server? I cannot find out where is the problem.
> >> Thankyou
> >> 
> >> 
> >> 
> >> 
> >> -- Best Regards,Sufeng NiuECASP lab, ECE department, Illinois Institute
> >> of
> >> TechnologyTel: 312-731-7219[4]
> >> 
> >> 
> >> --------[1] _ http://iocfccd1.aps.anl.gov_
> >> proxy%3A0%3A1 at iocfccd3.aps.anl.gov [13]
> >> http://iocfccd3.aps.anl.gov [8]
> >> 312-731-7219[10]
> >> http://lists.mpich.org/pipermail/discuss/attachments/20130621/01b37902/
> >> attachment-0001.html[14] >
> >> 
> >> ------------------------------
> >> 
> >> Message: 3Date: Sat, 22 Jun 2013 12:49:10 -0400From: Jiri Simsa
> >> < jsimsa at cs.cmu.edu [15]>To: discuss at mpich.org [2]
> >> CAHs9ut-_6W6SOHTJ_rD+shQ76bo4cTCuFVAy1f9x-
> >> J0gioakHg at mail.gmail.com [16]>Content-Type: text/plain;
> >> charset="iso-8859-1"
> >> 
> >> Hi,
> >> -------------- next part --------------
> >> An HTML attachment was scrubbed...
> >> URL: <
> >> http://lists.mpich.org/pipermail/discuss/attachments/20130623/a6104115/at
> >> ta
> >> chment.html >
> >> 
> >> ------------------------------
> >> 
> >> _______________________________________________
> >> discuss mailing list
> >> discuss at mpich.org
> >> https://lists.mpich.org/mailman/listinfo/discuss
> >> 
> >> End of discuss Digest, Vol 8, Issue 39
> >> **************************************
> > 
> > _______________________________________________
> > discuss mailing list     discuss at mpich.org
> > To manage subscription options or unsubscribe:
> > https://lists.mpich.org/mailman/listinfo/discuss
-- 
Antonio J. Peña
Postdoctoral Appointee
Mathematics and Computer Science Division
Argonne National Laboratory
9700 South Cass Avenue, Bldg. 240, Of. 3148
Argonne, IL 60439-4847
(+1) 630-252-7928
apenya at mcs.anl.gov

More information about the discuss mailing list