November 2012 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] determinism in collectives
by Jeff Hammond 12 Nov '12

12 Nov '12

We are primarily interested in MPICH2, as the motivation for this question relates to a particular supercomputer that uses MPICH2 and, in particular, the MPICH2 implementation of MPI_Reduce for (comm,type,op) of interest to us. Jeff On Mon, Nov 12, 2012 at 9:35 PM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > > The MPI standard does not guarantee this. For example, an > implementation could use topology-aware reductions or an MPI_ANY_SOURCE > while waiting for messages from peers making the order in which the > reduction operations are applied different for each run. > > -- Pavan > > On 11/12/2012 09:00 PM US Central Time, Nichols A. Romero wrote: >> Hi, >> >> I apologize if this has been asked a trillion times, if its documented somewhere just point it out to me online. >> >> Are collective operations, such as sums, using MPI_Reduce deterministic? In other words, if I run MPI_Reduce N times on P tasks, will the sums always occur in the same order. Thus it would be reproducible to the expected precision. >> > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [mpich-discuss] determinism in collectives
by Jed Brown 12 Nov '12

12 Nov '12

However, the standard does contain the following (MPI-3.0 page 175) *Advice to implementors.* *It is strongly recommended that MPI_REDUCE be im-* *plemented so that the same result be obtained whenever the function is applied on* *the same arguments, appearing in the same order. Note that this may prevent op-* *timizations that take advantage of the physical location of ranks. (End of advice to* *implementors.)* It would be useful to have a table of which implementations adhere to this advice. On Mon, Nov 12, 2012 at 9:35 PM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > The MPI standard does not guarantee this. For example, an > implementation could use topology-aware reductions or an MPI_ANY_SOURCE > while waiting for messages from peers making the order in which the > reduction operations are applied different for each run. >

1 0

Re: [mpich-discuss] determinism in collectives
by Pavan Balaji 12 Nov '12

12 Nov '12

The MPI standard does not guarantee this. For example, an implementation could use topology-aware reductions or an MPI_ANY_SOURCE while waiting for messages from peers making the order in which the reduction operations are applied different for each run. -- Pavan On 11/12/2012 09:00 PM US Central Time, Nichols A. Romero wrote: > Hi, > > I apologize if this has been asked a trillion times, if its documented somewhere just point it out to me online. > > Are collective operations, such as sums, using MPI_Reduce deterministic? In other words, if I run MPI_Reduce N times on P tasks, will the sums always occur in the same order. Thus it would be reproducible to the expected precision. > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

determinism in collectives
by Nichols A. Romero 12 Nov '12

12 Nov '12

Hi, I apologize if this has been asked a trillion times, if its documented somewhere just point it out to me online. Are collective operations, such as sums, using MPI_Reduce deterministic? In other words, if I run MPI_Reduce N times on P tasks, will the sums always occur in the same order. Thus it would be reproducible to the expected precision. -- Nichols A. Romero, Ph.D. Argonne Leadership Computing Facility Argonne National Laboratory Building 240 Room 2-127 9700 South Cass Avenue Argonne, IL 60490 (630) 252-3441

1 0

Mailing list migration announcement
by Pavan Balaji 12 Nov '12

12 Nov '12

Dear all, As a part of our preparation to move to the http://www.mpich.org domain, we have renamed our mailing lists as follows: mpich-announce(a)mcs.anl.gov is now called announce(a)mpich.org mpich-discuss(a)mcs.anl.gov is now called discuss(a)mpich.org mpich2-dev(a)mcs.anl.gov is now called devel(a)mpich.org We have migrated the users in each of these mailing lists to the new lists, including all your settings (such as digest mode, for example). However, we apologize if this causes discrepancy for any mail filters that you might have setup. The old mailing lists will still have active, but will forward all emails to the new lists. The website itself is offline for the moment (the old website http://www.mcs.anl.gov/research/projects/mpich2 is still available). The website will go live tomorrow. Regards, -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0