- mpiwg-io - mailman3.cels.anl.gov

Re: [MPI3-IO] Non blocking collective I/O ticket
by Mohamad Chaarawi 21 Dec '11

21 Dec '11

OK, I updated the draft with Rajeev's suggestions and uploaded it to the ticket. https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273 Thanks, Mohamad On 12/21/2011 11:11 AM, William Gropp wrote: > I like Rajeev's suggestion. > > Bill > > On Dec 20, 2011, at 2:33 PM, Rajeev Thakur wrote: > >>>> I was wondering if there is a way to forward reference this section >>>> somewhere earlier since it appears so late in the chapter. >>> >>> I was debating where should I have it. The other place was earlier >>> on page 447 line 12 (After Data Access Conventions). But then I saw >>> a section for collective I/O operation in the consistency semantics >>> Section, I put there. Do you think it would be better if we move it >>> up to the Data Access Routines subsection? Do you have another >>> suggestion? >> >> Or perhaps in the definition of the first nonblocking collective, >> there could be a forward reference to that section. For example, >> >> "MPI_FILE_IREAD_AT_ALL is a nonblocking version of >> MPI_FILE_READ_AT_ALL. See Section 13.6.5 on pg xxx for semantics of >> nonblocking collective file operations." >> >> Rajeev >> >> >> _______________________________________________ >> MPI3-IO mailing list >> MPI3-IO(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io > > William Gropp > Director, Parallel Computing Institute > Deputy Director for Research > Institute for Advanced Computing Applications and Technologies > Paul and Cynthia Saylor Professor of Computer Science > University of Illinois Urbana-Champaign > > > > _______________________________________________ > MPI3-IO mailing list > MPI3-IO(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by William Gropp 21 Dec '11

21 Dec '11

I like Rajeev's suggestion. Bill On Dec 20, 2011, at 2:33 PM, Rajeev Thakur wrote: >>> I was wondering if there is a way to forward reference this >>> section somewhere earlier since it appears so late in the chapter. >> >> I was debating where should I have it. The other place was earlier >> on page 447 line 12 (After Data Access Conventions). But then I saw >> a section for collective I/O operation in the consistency semantics >> Section, I put there. Do you think it would be better if we move it >> up to the Data Access Routines subsection? Do you have another >> suggestion? > > Or perhaps in the definition of the first nonblocking collective, > there could be a forward reference to that section. For example, > > "MPI_FILE_IREAD_AT_ALL is a nonblocking version of > MPI_FILE_READ_AT_ALL. See Section 13.6.5 on pg xxx for semantics of > nonblocking collective file operations." > > Rajeev > > > _______________________________________________ > MPI3-IO mailing list > MPI3-IO(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by Rajeev Thakur 20 Dec '11

20 Dec '11

>> I was wondering if there is a way to forward reference this section somewhere earlier since it appears so late in the chapter. > > I was debating where should I have it. The other place was earlier on page 447 line 12 (After Data Access Conventions). But then I saw a section for collective I/O operation in the consistency semantics Section, I put there. Do you think it would be better if we move it up to the Data Access Routines subsection? Do you have another suggestion? Or perhaps in the definition of the first nonblocking collective, there could be a forward reference to that section. For example, "MPI_FILE_IREAD_AT_ALL is a nonblocking version of MPI_FILE_READ_AT_ALL. See Section 13.6.5 on pg xxx for semantics of nonblocking collective file operations." Rajeev

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by Mohamad Chaarawi 19 Dec '11

19 Dec '11

Hi Rajeev, On 12/16/2011 02:20 PM, Rajeev Thakur wrote: > Instead of all input arguments, I would just say input buffers since there are no other array arguments and datatypes are refcounted. So the sentence could say > > "Input buffers should not be modified and output buffers should not be accessed before the completion call." fixed. > > I was wondering if there is a way to forward reference this section somewhere earlier since it appears so late in the chapter. I was debating where should I have it. The other place was earlier on page 447 line 12 (After Data Access Conventions). But then I saw a section for collective I/O operation in the consistency semantics Section, I put there. Do you think it would be better if we move it up to the Data Access Routines subsection? Do you have another suggestion? > > I also noticed that ln 2-3 on pg 446 needs to be marked as a change in MPI-3 (color). fixed. Thanks, Mohamad > > Rajeev

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by Rajeev Thakur 16 Dec '11

16 Dec '11

> All input arguments > should not be modified, and all output buffers should not be accessed before > the completion call. Instead of all input arguments, I would just say input buffers since there are no other array arguments and datatypes are refcounted. So the sentence could say "Input buffers should not be modified and output buffers should not be accessed before the completion call." I was wondering if there is a way to forward reference this section somewhere earlier since it appears so late in the chapter. I also noticed that ln 2-3 on pg 446 needs to be marked as a change in MPI-3 (color). Rajeev On Dec 14, 2011, at 10:17 AM, Mohamad Chaarawi wrote: > I updated the draft and attached it to the ticket with Fab's comments, which included moving the entire split collective section to the deprecated chapter, and emphasizing more the progress rules. > > I would like to get some feedback especially on the nonblocking Collective I/O section on page 484, which I updated to emphasize the progress rules which wasn't very clear in the first draft. > > Right now it reads as: > > \subsection{Nonblocking Collective File Operations} > %-------------------------------------- ; > \label{sec:io-semantics-nb-collective} > Nonblocking collective file operations are subject to the same restrictions as > blocking collective I/O operations. All processes, belonging to the > communicator that was used to open the file, must call collective I/O > operations (blocking and nonblocking) in the same order. In particular, once a > process calls a collective I/O operation, all other processes, belonging to > the communicator used to open the file, must eventually call the same > collective I/O operation, and no other collective I/O operation on the same > file handle in between. > > Nonblocking collective I/O operations do not match with blocking collective > I/O operations. Multiple nonblocking collective I/O operations can be > outstanding on a single file handle. Quality MPI implementations should be > able to support a large number of pending nonblocking I/O operations. > > All calls are local and return immediately, irrespective of the status of > other processes. The call initiates the operation which may progress > independently of any communication, computation, or I/O. The call returns a > request handle, which must be passed to a completion call. All input arguments > should not be modified, and all output buffers should not be accessed before > the completion call. The same progress rules described for nonblocking > collective operations apply for nonblocking collective I/O operations. For a > complete discussion, please refer to the semantics set forth in > Section~\ref{sec:nbcoll} on page~\pageref{sec:nbcoll}. > > > Thanks, > Mohamad > > On 12/13/2011 04:49 PM, Mohamad Chaarawi wrote: >> Hi All, >> >> I think enough people have joined to post to this list now.. >> Quincey will present ticket 273 (immediate non-blocking collective I/O operations) for a first reading at the January meeting. I attached an updated draft for IO chapter authors (or others) to review before sending it out on the general mailing list. >> Last time there was an issue with deprecating the split collective routines. Let me know if you have any comments. >> >> ticket url: >> https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273 >> >> Thanks, >> Mohamad >> >> >> >> _______________________________________________ >> MPI3-IO mailing list >> >> MPI3-IO(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io > > _______________________________________________ > MPI3-IO mailing list > MPI3-IO(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by Fab Tillier 14 Dec '11

14 Dec '11

Hi Mohamad, I read it over and it looks good to me. Cheers, -Fab Mohamad Chaarawi wrote on Wed, 14 Dec 2011 at 08:17:27 > > I updated the draft and attached it to the ticket with Fab's comments, > which included moving the entire split collective section to the > deprecated chapter, and emphasizing more the progress rules. > > I would like to get some feedback especially on the nonblocking > Collective I/O section on page 484, which I updated to emphasize the > progress rules which wasn't very clear in the first draft. > > Right now it reads as: > > \subsection{Nonblocking Collective File Operations} > %-------------------------------------- ; > \label{sec:io-semantics-nb-collective} Nonblocking collective file > operations are subject to the same restrictions as blocking collective > I/O operations. All processes, belonging to the communicator that was > used to open the file, must call collective I/O operations (blocking and > nonblocking) in the same order. In particular, once a process calls a > collective I/O operation, all other processes, belonging to the > communicator used to open the file, must eventually call the same > collective I/O operation, and no other collective I/O operation on the > same file handle in between. > > Nonblocking collective I/O operations do not match with blocking > collective I/O operations. Multiple nonblocking collective I/O > operations can be outstanding on a single file handle. Quality MPI > implementations should be able to support a large number of pending > nonblocking I/O operations. > > All calls are local and return immediately, irrespective of the status > of other processes. The call initiates the operation which may progress > independently of any communication, computation, or I/O. The call > returns a request handle, which must be passed to a completion call. All > input arguments should not be modified, and all output buffers should > not be accessed before the completion call. The same progress rules > described for nonblocking collective operations apply for nonblocking > collective I/O operations. For a complete discussion, please refer to > the semantics set forth in Section~\ref{sec:nbcoll} on > page~\pageref{sec:nbcoll}. > > > Thanks, > Mohamad > > On 12/13/2011 04:49 PM, Mohamad Chaarawi wrote: > Hi All, > > I think enough people have joined to post to this list now.. > Quincey will present ticket 273 (immediate non-blocking collective I/O > operations) for a first reading at the January meeting. I attached an > updated draft for IO chapter authors (or others) to review before > sending it out on the general mailing list. > Last time there was an issue with deprecating the split collective > routines. Let me know if you have any comments. > > ticket url: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273 > <https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273> > > Thanks, > Mohamad > > > > > > _______________________________________________ MPI3-IO mailing list > MPI3-IO(a)lists.mpi-forum.org <mailto:[email protected]> > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io > <http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io> > >

1 0

Re: [MPI3-IO] Non blocking collective I/O ticket
by Mohamad Chaarawi 14 Dec '11

14 Dec '11

I updated the draft and attached it to the ticket with Fab's comments, which included moving the entire split collective section to the deprecated chapter, and emphasizing more the progress rules. I would like to get some feedback especially on the nonblocking Collective I/O section on page 484, which I updated to emphasize the progress rules which wasn't very clear in the first draft. Right now it reads as: \subsection{Nonblocking Collective File Operations} %-------------------------------------- \label{sec:io-semantics-nb-collective} Nonblocking collective file operations are subject to the same restrictions as blocking collective I/O operations. All processes, belonging to the communicator that was used to open the file, must call collective I/O operations (blocking and nonblocking) in the same order. In particular, once a process calls a collective I/O operation, all other processes, belonging to the communicator used to open the file, must eventually call the same collective I/O operation, and no other collective I/O operation on the same file handle in between. Nonblocking collective I/O operations do not match with blocking collective I/O operations. Multiple nonblocking collective I/O operations can be outstanding on a single file handle. Quality MPI implementations should be able to support a large number of pending nonblocking I/O operations. All calls are local and return immediately, irrespective of the status of other processes. The call initiates the operation which may progress independently of any communication, computation, or I/O. The call returns a request handle, which must be passed to a completion call. All input arguments should not be modified, and all output buffers should not be accessed before the completion call. The same progress rules described for nonblocking collective operations apply for nonblocking collective I/O operations. For a complete discussion, please refer to the semantics set forth in Section~\ref{sec:nbcoll} on page~\pageref{sec:nbcoll}. Thanks, Mohamad On 12/13/2011 04:49 PM, Mohamad Chaarawi wrote: > Hi All, > > I think enough people have joined to post to this list now.. > Quincey will present ticket 273 (immediate non-blocking collective I/O > operations) for a first reading at the January meeting. I attached an > updated draft for IO chapter authors (or others) to review before > sending it out on the general mailing list. > Last time there was an issue with deprecating the split collective > routines. Let me know if you have any comments. > > ticket url: > https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273 > > Thanks, > Mohamad > > > > _______________________________________________ > MPI3-IO mailing list > MPI3-IO(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-io

1 0

[MPI3-IO] Non blocking collective I/O ticket
by Mohamad Chaarawi 13 Dec '11

13 Dec '11

Hi All, I think enough people have joined to post to this list now.. Quincey will present ticket 273 (immediate non-blocking collective I/O operations) for a first reading at the January meeting. I attached an updated draft for IO chapter authors (or others) to review before sending it out on the general mailing list. Last time there was an issue with deprecating the split collective routines. Let me know if you have any comments. ticket url: https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/273 Thanks, Mohamad

1 0