2.1 we do expect to have a somewhat short lifetime, depending on what happens with MPI 2.2 and 3.0. However, this is the base for future revisions of the standard, so I think that it is worth slipping this schedule a bit so that we have a firm base to build on. The document will never be perfect, and this is not what we should strive for, but we should strive to have it be in a form that someone can pick up the document and implement the standard correctly. I will also add that I have talked with some who have committed to do a major portion of work for 2.1 who simply do not have the cycles to do this for several weeks. I am in the same situation. Rich On 4/3/08 8:09 AM, "Rolf Rabenseifner" <rabenseifner_at_[hidden]> wrote:
Hi all,
yes, we need a decision as soon as possible.
But first, it may be necessary to remove a wrong argument: - The standard will have 6 month lifetime, not more! - MPI 2.1 has to major goals: - make Ballots 1-4 as part of the official standard, - reduce the number of documents to one. All rest is nice to have - not a major goal. Until next Friday, only secure changes to intoductional text parts should be done. More must be moved to MPI-2.2.
To have a monolithic MPI standard document can be added to the agenda of MPI-2.2.
Here the official Scope of Effort as defined in January (see protocol):
Scope of Effort: - Clarification to the MPI standards document, - resulting in a single document describing the full MPI 2.1 standard. - This includes merging of documents, text corrections, and added clarifying text.
All these three goals are achieved! Current task is only to clean up small things, mainly (March protocol):
Additional text-merge round with dedicated chapter authors. Goal: - Remove references to MPI-1 and MPI-2. - Substitute by MPI. - Necessary modifications to achieve this goal.
MPI-2.2 can do it better in 6 month. The MPI-2.2 for this can start about 3 weeks after April meeting.
I hope, you all can live with this decisions. (My work time is limited; this is another reason why I cannot accept a delay in the delivery of MPI-2.1)
Bronis and George,
my general statements do not answer you initial question: If you decide to move parts from Chap.11 to Chap.7, then you both mus discuss this. You both are responsible for these chapters. And you should first convince your reviewers: - Chap. 7: Rich, Jesper, Steve, Kannan, David, Bill - Chap.11: Bill and Rainer My recommendation: Express clearly which parts should be moved exactly to wich line (all based on page/line numbers as **printed** in Draft Apr. 1, 2008).
Jeff,
The idea of moving significant parts of Chap. 13 into Terms is absolutely against the spirit of how this document is written. The author tried to have only absolute necessary information before the fist MPI example on page 25 (In MPI-1, it was page 16). The MPI-2 Forum decided to put all the more detailed information into extra sections, and especially binding stuff at the end of the standard. I put the Profiling behind Binding (this may be a fault).
C and Fortran have the clear advantage that all constants have the same name as the language-independently defined name, e.g. MPI_COMM_WORLD, MPI_INT, MPI_SUCCESS... .
For MPI-2.1, I would keep it as it is. In MPI-2.2, there will be a longer discussion on C++ issues.
Best regards Rolf
On Thu, 3 Apr 2008 09:28:08 +0100 "Supalov, Alexander" <alexander.supalov_at_[hidden]> wrote:
Hi,
Whatever the decision will be, we need it now. We're close to the point of no return for the September deadline.
Best regards.
Alexander
-----Original Message----- From: Jesper Larsson Traeff [mailto:traff_at_it.neclab.eu] Sent: Thursday, April 03, 2008 9:26 AM To: Bronis R. de Supinski Cc: Supalov, Alexander; Rolf Rabenseifner; William Gropp; Tony Skjellum; Rich Graham; Adam Moody; Richard Treumann; George Bosilca; David Solt; Rajeev Thakur; Jeff Squyres; MPI 2.1 Mailing List Subject: Re: TODO - AUTHORS of MPI-2.1 - Workplan
Dear All,
I agree with the comments of Bronis about datatypes - functionality that belongs together should be together in the STANDARD! (which is what we are writing). To me it seems that this 2.1 is somewhat more than the "merge" it was optimistically intended to be. Since this, as Bronis says, will likely be THE STANDARD for years to come, and the entry point for many new users, I think it is important that we get it right the first time, and I believe this is possible (if not, we should abort the attempt). It's good to have strict deadlines, but let's not be fanatical
about that. I don't see any purpose whatsoever in putting out intermediate versions of 2.1 - as long as we are in limbo, 1.3 and 2.0 will work perfectly, they have all the ballots included
I will try to send comments on/updates of my chapters (these will, I think/hope, not need so much) tomorrow/monday
best regards
Jesper
All:
I thoroughly agree with the sentiment that many expressed that we are better off slipping the release of 2.1 two or four months more in order to get it right than hurrying to get it out the door. It is likely to be THE standard for a fairly long time and having it broken to begin with does not seem wise.
Bronis
On Wed, 2 Apr 2008, Supalov, Alexander wrote:
Hi,
We may want to reduce the number of top level sections in the
standard.
In my opinion that I expressed to Rolf a couple of weeks ago, quite a few sections, like the ones mentioned by Bronis, should rather belong to a big Miscellany chapter rather than figure up there by themselves. They may also be merged/reformed, too.
However, this is going to be a bigger change than envisioned originally. I wonder whether we should decide right away whether we can afford this now without slipping the planned MPI-2.1 delivery in September. A possible way would be to fix small things now and do second pass of major edits/reshuffling in MPI-3, or MPI-2.2, for that matter.
Best regards.
Alexander
-----Original Message----- From: Bronis R. de Supinski [mailto:bronis_at_[hidden]] Sent: Wednesday, April 02, 2008 11:15 PM To: Rolf Rabenseifner Cc: William Gropp; Tony Skjellum; Rich Graham; Adam Moody; Richard Treumann; Jespar Larsson Traeff; George Bosilca; David Solt; Rajeev Thakur; Jeff Squyres; Supalov, Alexander; MPI 2.1 Mailing List Subject: Re: TODO - AUTHORS of MPI-2.1 - Workplan
Rolf:
I have not gone through Chapter 11 thoroughly yet but I have already noticed some major changes that I would suggest.
First Section 11.6: "Decoding a Datatype" seems out of place in the merged document. It would make sense to me to have all of the datatype functions together. If we don't make datatypes a separate chapter, then this section (11.6) should be moved into chapter 3, near section 3.12: "Derived Datatypes". Ideally, it would be merged into that section since it is clearly part of that functionality.
Second, Section 11.5: "Error Classes, Error Codes, and Error Handlers" is strongly related to Section 7.3: "Error Handling", and Section 7.4: "Error Codes and Classes". Clearly, these sections should be merged. In fact, this overlap makes me ask why have we not merged Chapters 7 and 11? What is the difference between "Environmental Management" and "External Interfaces"?
I think this question needs to be resolved before I go further on working on Chapter 11. Perhaps George and I should coordinate merging these chapters after I coordinate moving the datatype decoding functionality into chapter 3 with Rich...
Bronis
On Wed, 2 Apr 2008, Bronis R. de Supinski wrote:
Rolf:
I have finished editing chapter 14: Profiling. I have attached the modified prof.tex since I do not yet have write access. More importantly, either there is a problem with the 2.1 macros or I am doing something wrong (I think it is the first but I am not LaTex fluent enough to be certain). Specifically, the first letter of the word following the end markers does not appear in the generated PDF. I would appreciate it if you could look into which it is.
Another observation is that the macro does not work around empty sections. In particular, I tried to mark the section that I moved with the macros, with the original text commented out. I tried to do the same for some MPI-1.0 text that was contradictory to or redundant with the text that I moved. Having those empty macro regions caused the compile to fail. It might be nice to have some way to flag the deleted text, with the convention, at least for now, of just commenting it out and not to delete it entirely from the file.
Let me know if I am doing things right here and then I will move on to fixing chapter 11: External Interfaces. Thanks,
Bronis
On Wed, 2 Apr 2008, Rolf Rabenseifner wrote:
Dear Bill, Tony, Rich, Adam, Dick, Jesper, George, David,
Bronis,
Rajeev, Jeff, and Alexander,
you are responsible for one ore more chapters of MPI-2.1 until
the
meeting.
It is a tough schedule: -----------------------
* Tues, April 1, my ROUND ONE is finished and the write token is logically passed to all chapter authors: - You have read access to the source (via SVN) and the pdf. - You have write access to your local SVN copy. - You will get write access to the SVN in a few days. * Fri., April 11, chapter authors have finished ROUND TWO. - This is a hard deadline (because I've to go on travel) * Sat., April 12, I will produce mpi-report.pdf as basis for review (I cannot do later, because April 14-17, I'm on travel) * Mon.-Thu., April 14-17, strong review by the reviewer group * Fri., April 18, reviews - if necessary must be included - by the chapter authors * Sat., April 19, I will produce final mpi-report.pdf which is basis for - final reviews through the reviewer group - official reading at the April 28-30, 2008 meeting. (I cannot do later because I'm on travel April 20-26)
The goals of your work as chapter author: -----------------------------------------
- Remove references to MPI-1 and MPI-2. - Substitute by MPI. - Necessary modifications to achieve this goal.
With this, we should have a single MPI-2.1 standard that does not "know" the MPI-1 or MPI-2 history of individual functions.
** This should be mainly a task in the area of chapter-introductions. ** Please, never change the wording of function definitions.
Exceptions: - There are routines that are deprecated and that are already referenced in a consistent way. My recommendation: Current wording: There are *new* function, and existing are now deprecated. Proposal: There are functions, and there exist also deprecated functions with (nearly) same functionality but deprecated due to some lack in the bindings (or functionality) - There is history information in the frontmatter. - there is a change-log annex with limited memory (only previous version)
Technical editing rule: -----------------------
You must identify all of your changes:
a) new/modified wording - you are highlighting your new/modified wording with:
\mpiiidotiMergeNEWforSINGLEbegin% MPI-2.1 round-two - begin of modification .... your modified / new wording .... \mpiiidotiMergeNEWforSINGLEendI% MPI-2.1 round-two - end of modification
b) Moved paragraphs, sentences, ... - you are highlighting the first word of the moved text with:
\mpiiidotiMergeNEWforSINGLEbegin% MPI-2.1 round-two - begin of text-move First-word \mpiiidotiMergeNEWforSINGLEendI% MPI-2.1 round-two - ongoing text-move ... rest of the moved text.... % MPI-2.1 round-two - end of text-move
Caution: These macros work only in black parts of the text.
In blue parts you must substitute \mpiiidotiMergeNEWforSINGLEendI% by \mpiiidotiMergeNEWforSINGLEendII%
The highlighting is done with red color. (There are only a few other sentences from me magenta. Therefore red should be the best to highlight the round-two modifications.)
Rsponsibilities: ----------------
* Frontmatter mpi-report.tex Bill Gropp * Acknowledgements credits.tex Rich Graham(text) + Rolf(emails-auth.) * 1. Introduction intro.tex Bill Gropp - 2. Terms terms-2.tex Tony Skjellum * 3. Point-to-point pt2pt.tex Rich Graham * 4. Collectives coll.tex Adam Moody * 5. Groups, etc. context.tex Dick Treumann - 6. Toplogies topol.tex Jesper Traeff * 7. Environment inquiry.tex George Bosilca - 8. Miscellany misc-2.tex Jesper Traeff - 9. Process Crea... dynamic-2.tex David Solt - 10. One-sided Comm one-sided-2.tex Jepser Traeff - 11. External Int. ei-2.tex Bronis de Supinski - 12. IO io-2.tex Rajeev Thakur - 13. Lang.Binding binding-2.tex Jeff Squyres - 14. Profiling prof.tex Bronis de Supinski - 15. Deprecated deprecated.tex Rolf Rabenseifner - Annex A Lang.Bind. appLang*.tex, MAKE-APPLANG Alexander Supalov - Annex B Change-log changes.tex Rolf Rabenseifner - Bibliography refs.bib Bill Gropp - Index MAKE-FUNC-INDEX Rolf Rabenseifner
Legende: - Chapter is mainly taken from MPI-1.1 or MPI-2.0, only small work expected * There was a significant merge, or new text, or ..., more work expected
I wish you a good start and good luck with your chapter.
Best regards Rolf
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring
On Wed, Apr 02, 2008 at 02:26:13PM -0700, Bronis R. de Supinski wrote: 30)
Intel GmbH Dornacher Strasse 1 85622 Feldkirchen/Muenchen Germany Sitz der Gesellschaft: Feldkirchen bei Muenchen Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer Registergericht: Muenchen HRB 47456 Ust.-IdNr. VAT Registration No.: DE129385895 Citibank Frankfurt (BLZ 502 109 00) 600119052
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
Intel GmbH Dornacher Strasse 1 85622 Feldkirchen/Muenchen Germany Sitz der Gesellschaft: Feldkirchen bei Muenchen Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer Registergericht: Muenchen HRB 47456 Ust.-IdNr. VAT Registration No.: DE129385895 Citibank Frankfurt (BLZ 502 109 00) 600119052
This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring 30) _______________________________________________ mpi-21 mailing list mpi-21_at_[hidden] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-21
*