May 2012 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] Notes on discussions about 3.0 / 3.1?
by Jeff Squyres 31 May '12

31 May '12

This probably revives the "MPI-next" discussion. I think we hadn't figured out what the "next" version of MPI would be, and punted the conversation down the road so that we could concentrate on 3.0. FWIW, I see at least a few different directions of discussion: - what types of things are going into the next MPI release: bug fixes, new features, or both? - whether to keep the same convention that we did in MPI 2.x (i.e., only bug fixes into 3.x) - sub topic: should we do MPI 3.1 (for bugs) and MPI 4.0 (for new things)? --> this would be a lot of effort for 4 meetings/year, but it's worth discussing and making a concrete decision - what to call the next release On May 31, 2012, at 6:02 PM, Jeff Hammond wrote: > Perhaps since MPI 3.0 got everything right that it attempted to do :-) > then we can skip 3.1 and move onto to 4.0, the most nonblocking-est, > fault-tolerantest MPI ever! > > Jeff > > On Thu, May 31, 2012 at 4:59 PM, Jeff Squyres <jsquyres(a)cisco.com> wrote: >> On May 31, 2012, at 4:58 PM, Jeff Hammond wrote: >> >>> It would be useful to know what is not possible for 3.1. For example, >>> I've been told that NBC could not have been done as MPI 2.3 because >>> they were a major change and that can't happen on minor version >>> releases. Does that mean anything drastically new in form (e.g. >>> active messages not derived from MPI_Accumulate) are off the table and >>> have to wait for 4.0? >> >> FWIW, the same question occurred to me, especially with references earlier today that FT and/or iFile could come back in MPI 3.1 (i.e., we explicitly did not allow new things in MPI 2.2.x). >> >>> Jeff >>> >>> On Thu, May 31, 2012 at 3:48 PM, Richard Graham <richardg(a)mellanox.com> wrote: >>>> Comments in line >>>> >>>> -----Original Message----- >>>> From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres >>>> Sent: Friday, June 01, 2012 4:21 AM >>>> To: MPI Forum list >>>> Subject: [Mpi-forum] Notes on discussions about 3.0 / 3.1? >>>> >>>> Did anyone take any notes on the following two agenda items from the meeting this past week: >>>> >>>> - MPI 3.1 - initial plans >>>> [rich] Not much discussion at this stage. The one thing I do remember is that we started to discuss voting, given that we are moving to a format of meeting once a quarter. One of the things suggested was that we have a day in between meetings for voting, doing this over the phone, but with attendance at the regular meetings continuing to determine voting elegibility. >>>> >>>> - MPI 3.0 - Draft standard >>>> [rich] (1) Push to get all the voted in tickets into the draft standard by JUNE 15th. This requires ticket owners and chapter committees to work together. Chapter authors that don't participate would be replaced by others that will help get the revised text into the draft standard. (2) The July Chicago meeting will be devoted solely for getting the draft standard into final form. Last second votes will take place on Monday, with the rest of the time dedicated to chapter committee work. (3) The draft standard will be posted for public comment after the meeting, giving people time to comment before the final votes in Vienna. >>>> >>>> Rich >>>> >>>> >>>> I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. >>>> >>>> Thanks. >>>> >>>> -- >>>> Jeff Squyres >>>> jsquyres(a)cisco.com >>>> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >>>> >>>> >>>> _______________________________________________ >>>> mpi-forum mailing list >>>> mpi-forum(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>>> >>>> _______________________________________________ >>>> mpi-forum mailing list >>>> mpi-forum(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >>> >>> >>> -- >>> 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 >>> >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> >> >> -- >> Jeff Squyres >> jsquyres(a)cisco.com >> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >> >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > > -- > 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 > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] Notes on discussions about 3.0 / 3.1?
by Jeff Hammond 31 May '12

31 May '12

Perhaps since MPI 3.0 got everything right that it attempted to do :-) then we can skip 3.1 and move onto to 4.0, the most nonblocking-est, fault-tolerantest MPI ever! Jeff On Thu, May 31, 2012 at 4:59 PM, Jeff Squyres <jsquyres(a)cisco.com> wrote: > On May 31, 2012, at 4:58 PM, Jeff Hammond wrote: > >> It would be useful to know what is not possible for 3.1. ï¿½For example, >> I've been told that NBC could not have been done as MPI 2.3 because >> they were a major change and that can't happen on minor version >> releases. ï¿½Does that mean anything drastically new in form (e.g. >> active messages not derived from MPI_Accumulate) are off the table and >> have to wait for 4.0? > > FWIW, the same question occurred to me, especially with references earlier today that FT and/or iFile could come back in MPI 3.1 (i.e., we explicitly did not allow new things in MPI 2.2.x). > >> Jeff >> >> On Thu, May 31, 2012 at 3:48 PM, Richard Graham <richardg(a)mellanox.com> wrote: >>> Comments in line >>> >>> -----Original Message----- >>> From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres >>> Sent: Friday, June 01, 2012 4:21 AM >>> To: MPI Forum list >>> Subject: [Mpi-forum] Notes on discussions about 3.0 / 3.1? >>> >>> Did anyone take any notes on the following two agenda items from the meeting this past week: >>> >>> - MPI 3.1 - initial plans >>> [rich] Not much discussion at this stage. ï¿½The one thing I do remember is that we started to discuss voting, given that we are moving to a format of meeting once a quarter. ï¿½One of the things suggested was that we have a day in ï¿½between meetings for voting, doing this over the phone, but with attendance at the regular meetings continuing to determine voting elegibility. >>> >>> - MPI 3.0 - Draft standard >>> [rich] (1) ï¿½Push to get all the voted in tickets into the draft standard by JUNE 15th. ï¿½This requires ticket owners and chapter committees to work together. ï¿½Chapter authors that don't participate would be replaced by others that will help get the revised text into the draft standard. (2) The July Chicago meeting will be devoted solely for getting the draft standard into final form. ï¿½Last second votes will take place on Monday, with the rest of the time dedicated to chapter committee work. ï¿½(3) The draft standard will be posted for public comment after the meeting, giving people time to comment before the final votes in Vienna. >>> >>> Rich >>> >>> >>> I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. >>> >>> Thanks. >>> >>> -- >>> Jeff Squyres >>> jsquyres(a)cisco.com >>> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >>> >>> >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> >> >> >> -- >> 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 >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- 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: [Mpi-forum] Notes on discussions about 3.0 / 3.1?
by Jeff Squyres 31 May '12

31 May '12

On May 31, 2012, at 4:58 PM, Jeff Hammond wrote: > It would be useful to know what is not possible for 3.1. For example, > I've been told that NBC could not have been done as MPI 2.3 because > they were a major change and that can't happen on minor version > releases. Does that mean anything drastically new in form (e.g. > active messages not derived from MPI_Accumulate) are off the table and > have to wait for 4.0? FWIW, the same question occurred to me, especially with references earlier today that FT and/or iFile could come back in MPI 3.1 (i.e., we explicitly did not allow new things in MPI 2.2.x). > Jeff > > On Thu, May 31, 2012 at 3:48 PM, Richard Graham <richardg(a)mellanox.com> wrote: >> Comments in line >> >> -----Original Message----- >> From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres >> Sent: Friday, June 01, 2012 4:21 AM >> To: MPI Forum list >> Subject: [Mpi-forum] Notes on discussions about 3.0 / 3.1? >> >> Did anyone take any notes on the following two agenda items from the meeting this past week: >> >> - MPI 3.1 - initial plans >> [rich] Not much discussion at this stage. The one thing I do remember is that we started to discuss voting, given that we are moving to a format of meeting once a quarter. One of the things suggested was that we have a day in between meetings for voting, doing this over the phone, but with attendance at the regular meetings continuing to determine voting elegibility. >> >> - MPI 3.0 - Draft standard >> [rich] (1) Push to get all the voted in tickets into the draft standard by JUNE 15th. This requires ticket owners and chapter committees to work together. Chapter authors that don't participate would be replaced by others that will help get the revised text into the draft standard. (2) The July Chicago meeting will be devoted solely for getting the draft standard into final form. Last second votes will take place on Monday, with the rest of the time dedicated to chapter committee work. (3) The draft standard will be posted for public comment after the meeting, giving people time to comment before the final votes in Vienna. >> >> Rich >> >> >> I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. >> >> Thanks. >> >> -- >> Jeff Squyres >> jsquyres(a)cisco.com >> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ >> >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > > -- > 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 > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] Collection of voting points
by Jeff Squyres 31 May '12

31 May '12

On May 31, 2012, at 2:14 PM, Richard Graham wrote: > The posted voting document was the result of a forum discussion to clarify the voting rules for MPI 3.0 - e.g., an attempt to go beyond the few lines that were written before, w/o changing the rules that have been in place since MPI-1, just clarifying them, for those that want to see a detailed and explicit set of rules. Understood, but I think what we're discovering through these threads is that this document is inadequate -- there are multiple undefined terms that have created the current controversy. Plus, there was no formal paper trail of the acceptance of this document (e.g., voting). Per my points from the original mail, at least the following terms still need to be defined: - whether "abstain" counts as a "no" or not - what exactly "present" means in the phrase about simple majority - what exactly constitutes a quorum These need to be decided before we vote again in July. I understand how some things are better left as "fuzzily defined" -- but when it comes to the voting rules, there can be no interpretation wiggle room... or we'll just have this exact same controversy again sometime in the future. Finally, I have come around to the view that I think we have to accept the ballot results as stated from this past meeting. There's just no sane way to change what has happened without opening multiple other cans of worms. > Rich > > -----Original Message----- > From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres > Sent: Thursday, May 31, 2012 9:39 PM > To: MPI Forum list > Subject: [Mpi-forum] Collection of voting points > > BTW, I want to stress that I think that this is good, healthy debate for the Forum. Thank you all for taking the time to add your voice to the discussion. > > In reading the thread that continued last night, I wish to make the following points: > > 1. What is currently being referred to as "the bylaws" is a document that Rich checked into Subversion on May 9, 2011 (i.e., about a year ago). I do recall that there was some discussion about this document before it was posted, but I do not believe that it received the same level of scrutiny as MPI Forum document text, for example. I do not believe that this document had 2 binding votes, either (someone please correct me if I am wrong). > > *** Meaning: this is a fairly recent document. > > 2. The first votes I recorded (for MPI-2.2 stuff) was in March, 2008 (see http://meetings.mpi-forum.org/Meeting_details.php and https://svn.mpi-forum.org/svn/mpi-forum-docs/trunk/meetings/2008-03-march/v…) The first MPI-3 vote I see was for NBC in September, 2010. > > All of these votes were recorded with a "yes votes > no votes" computation. However, as has been pointed out, before the May 2012 meeting, I don't think we've had many (any?) close-enough official votes such that "yes>no" vs. "yes>50%" would have mattered. > > *** Meaning: there were many votes before the procedures document was published. When paired with the fact that the document was not voted on, referring to that document as "bylaws", I think, is therefore a little misleading. > > 3. In every single vote record, there are separate columns for yes, no, and abstain (and miss). > > *** Meaning: This concept of a separate "abstain" result is exactly why I used the original "yes votes > no votes" computation. The existence of "abstain" clearly indicates to me that the Forum intended "abstain" to be different than "no". Note, too, that "abstain" existed in MPI-2 (I wasn't part of MPI-1). > > 4. Starting in June, 2009, all ballot result web pages contain the following text: > > NOTE: "Missed" counts the same as "abstain". > > *** Meaning: This notice did not appear on the 6+1 prior ballot result web pages. I do not remember what happened here, but this *suggests* that there was some discussion in the June 2009 meeting about the exact meaning of "missed". > > My own belief is that "missed" is just an annotation for inside the forum -- it helps us remember, "Oh, yeah, Joe's plane was late and he missed that vote." Or, "Oh, yeah, Bob must have stepped out of the room for that vote." > > 5. Regardless of all of the above, there has never been a definition of "present" in the quoted definition of voting: > > a simple majority is defined as a simple majority > of those present and eligible to vote. > > Does "present" mean "registered to attend the meeting", or "registered to attend the meeting AND physically present in the room at the time of the vote"? > > *** Meaning: are we saying that there is difference between saying "abstain" and walking out of the room? That seems like a specious distinction, to me. > > *** Additionally: this specifically calls into question the meaning of the "missed" votes (note that there were no "missed" votes in the Japan meeting, but this point also needs to be clarified). > > 6. A conspicuous definition that is also absent -- and we have discussed its absence many times -- is that of "quorum". How many voting-eligible organizations are required to constitute critical mass such that a binding vote can take place? > > Again, as someone correctly pointed out, this was not a factor for the Japan meeting. Here's some stats to show that the Japan votes were representative of the overall Forum: > > - May 2012 (Japan): 16 orgs voting > - March 2012 (Chicago): 17 orgs voting > - Jan 2012 (San Jose): 16 orgs voting > - ... I didn't bother to go back further > > *** Meaning: If we're quibbling over the voting rules, the term "quorum" also needs to be defined. > > === OVERALL SUMMARY: I personally don't have a strong opinion on which set of definitions we end up using. Indeed, Bronis (and Torsten, in IM's with me) make good points about needing a large number of "yes" votes to pass. So this is a good debate to have: if we want to abolish "abstain" (and its synonym, "miss"), then fine -- let's debate and make a conscious, Forum-wide decision to do so. > > === My objection is that the voting records CLEARLY show that we have treated abstains different than "no"; this is a tradition that extends back to MPI-2. I object to the sudden changing of the ballot rules as I, the secretary, have understood, recorded, and transparently published the ballot results. It seems to be a problem that people were not paying attention to *the last 4 years of ballots*, but that does not mean that the discernment of ballots can change with no discussion by the Forum. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] Notes on discussions about 3.0 / 3.1?
by Jeff Hammond 31 May '12

31 May '12

It would be useful to know what is not possible for 3.1. For example, I've been told that NBC could not have been done as MPI 2.3 because they were a major change and that can't happen on minor version releases. Does that mean anything drastically new in form (e.g. active messages not derived from MPI_Accumulate) are off the table and have to wait for 4.0? Jeff On Thu, May 31, 2012 at 3:48 PM, Richard Graham <richardg(a)mellanox.com> wrote: > Comments in line > > -----Original Message----- > From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres > Sent: Friday, June 01, 2012 4:21 AM > To: MPI Forum list > Subject: [Mpi-forum] Notes on discussions about 3.0 / 3.1? > > Did anyone take any notes on the following two agenda items from the meeting this past week: > > - MPI 3.1 - initial plans > [rich] Not much discussion at this stage. ï¿½The one thing I do remember is that we started to discuss voting, given that we are moving to a format of meeting once a quarter. ï¿½One of the things suggested was that we have a day in ï¿½between meetings for voting, doing this over the phone, but with attendance at the regular meetings continuing to determine voting elegibility. > > - MPI 3.0 - Draft standard > [rich] (1) ï¿½Push to get all the voted in tickets into the draft standard by JUNE 15th. ï¿½This requires ticket owners and chapter committees to work together. ï¿½Chapter authors that don't participate would be replaced by others that will help get the revised text into the draft standard. (2) The July Chicago meeting will be devoted solely for getting the draft standard into final form. ï¿½Last second votes will take place on Monday, with the rest of the time dedicated to chapter committee work. ï¿½(3) The draft standard will be posted for public comment after the meeting, giving people time to comment before the final votes in Vienna. > > Rich > > > I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. > > Thanks. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- 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: [Mpi-forum] Notes on discussions about 3.0 / 3.1?
by Richard Graham 31 May '12

31 May '12

Comments in line -----Original Message----- From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres Sent: Friday, June 01, 2012 4:21 AM To: MPI Forum list Subject: [Mpi-forum] Notes on discussions about 3.0 / 3.1? Did anyone take any notes on the following two agenda items from the meeting this past week: - MPI 3.1 - initial plans [rich] Not much discussion at this stage. The one thing I do remember is that we started to discuss voting, given that we are moving to a format of meeting once a quarter. One of the things suggested was that we have a day in between meetings for voting, doing this over the phone, but with attendance at the regular meetings continuing to determine voting elegibility. - MPI 3.0 - Draft standard [rich] (1) Push to get all the voted in tickets into the draft standard by JUNE 15th. This requires ticket owners and chapter committees to work together. Chapter authors that don't participate would be replaced by others that will help get the revised text into the draft standard. (2) The July Chicago meeting will be devoted solely for getting the draft standard into final form. Last second votes will take place on Monday, with the rest of the time dedicated to chapter committee work. (3) The draft standard will be posted for public comment after the meeting, giving people time to comment before the final votes in Vienna. Rich I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. Thanks. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Notes on discussions about 3.0 / 3.1?
by Jeff Squyres 31 May '12

31 May '12

Did anyone take any notes on the following two agenda items from the meeting this past week: - MPI 3.1 - initial plans - MPI 3.0 - Draft standard I wasn't at the meeting, but would love to hear what the general discussion was around these two topics. Thanks. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] Collection of voting points
by Richard Graham 31 May '12

31 May '12

The posted voting document was the result of a forum discussion to clarify the voting rules for MPI 3.0 - e.g., an attempt to go beyond the few lines that were written before, w/o changing the rules that have been in place since MPI-1, just clarifying them, for those that want to see a detailed and explicit set of rules. Rich -----Original Message----- From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Jeff Squyres Sent: Thursday, May 31, 2012 9:39 PM To: MPI Forum list Subject: [Mpi-forum] Collection of voting points BTW, I want to stress that I think that this is good, healthy debate for the Forum. Thank you all for taking the time to add your voice to the discussion. In reading the thread that continued last night, I wish to make the following points: 1. What is currently being referred to as "the bylaws" is a document that Rich checked into Subversion on May 9, 2011 (i.e., about a year ago). I do recall that there was some discussion about this document before it was posted, but I do not believe that it received the same level of scrutiny as MPI Forum document text, for example. I do not believe that this document had 2 binding votes, either (someone please correct me if I am wrong). *** Meaning: this is a fairly recent document. 2. The first votes I recorded (for MPI-2.2 stuff) was in March, 2008 (see http://meetings.mpi-forum.org/Meeting_details.php and https://svn.mpi-forum.org/svn/mpi-forum-docs/trunk/meetings/2008-03-march/v…) The first MPI-3 vote I see was for NBC in September, 2010. All of these votes were recorded with a "yes votes > no votes" computation. However, as has been pointed out, before the May 2012 meeting, I don't think we've had many (any?) close-enough official votes such that "yes>no" vs. "yes>50%" would have mattered. *** Meaning: there were many votes before the procedures document was published. When paired with the fact that the document was not voted on, referring to that document as "bylaws", I think, is therefore a little misleading. 3. In every single vote record, there are separate columns for yes, no, and abstain (and miss). *** Meaning: This concept of a separate "abstain" result is exactly why I used the original "yes votes > no votes" computation. The existence of "abstain" clearly indicates to me that the Forum intended "abstain" to be different than "no". Note, too, that "abstain" existed in MPI-2 (I wasn't part of MPI-1). 4. Starting in June, 2009, all ballot result web pages contain the following text: NOTE: "Missed" counts the same as "abstain". *** Meaning: This notice did not appear on the 6+1 prior ballot result web pages. I do not remember what happened here, but this *suggests* that there was some discussion in the June 2009 meeting about the exact meaning of "missed". My own belief is that "missed" is just an annotation for inside the forum -- it helps us remember, "Oh, yeah, Joe's plane was late and he missed that vote." Or, "Oh, yeah, Bob must have stepped out of the room for that vote." 5. Regardless of all of the above, there has never been a definition of "present" in the quoted definition of voting: a simple majority is defined as a simple majority of those present and eligible to vote. Does "present" mean "registered to attend the meeting", or "registered to attend the meeting AND physically present in the room at the time of the vote"? *** Meaning: are we saying that there is difference between saying "abstain" and walking out of the room? That seems like a specious distinction, to me. *** Additionally: this specifically calls into question the meaning of the "missed" votes (note that there were no "missed" votes in the Japan meeting, but this point also needs to be clarified). 6. A conspicuous definition that is also absent -- and we have discussed its absence many times -- is that of "quorum". How many voting-eligible organizations are required to constitute critical mass such that a binding vote can take place? Again, as someone correctly pointed out, this was not a factor for the Japan meeting. Here's some stats to show that the Japan votes were representative of the overall Forum: - May 2012 (Japan): 16 orgs voting - March 2012 (Chicago): 17 orgs voting - Jan 2012 (San Jose): 16 orgs voting - ... I didn't bother to go back further *** Meaning: If we're quibbling over the voting rules, the term "quorum" also needs to be defined. === OVERALL SUMMARY: I personally don't have a strong opinion on which set of definitions we end up using. Indeed, Bronis (and Torsten, in IM's with me) make good points about needing a large number of "yes" votes to pass. So this is a good debate to have: if we want to abolish "abstain" (and its synonym, "miss"), then fine -- let's debate and make a conscious, Forum-wide decision to do so. === My objection is that the voting records CLEARLY show that we have treated abstains different than "no"; this is a tradition that extends back to MPI-2. I object to the sudden changing of the ballot rules as I, the secretary, have understood, recorded, and transparently published the ballot results. It seems to be a problem that people were not paying attention to *the last 4 years of ballots*, but that does not mean that the discernment of ballots can change with no discussion by the Forum. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] Voting results
by David Solt 31 May '12

31 May '12

I want to clarify that I did object that the way we were handling votes at this last meeting was inconsistent with previous meetings, but those people who spoke up simply indicated that I was mistaken. I regret that I did not push the issue further, but there did not appear to me to be an intention to change how things were done. It appeared to me that as a group we were simply wrong about our understanding of the process. Thanks, Dave From: Jeff Hammond <jhammond(a)alcf.anl.gov> To: Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> Date: 05/30/2012 08:32 PM Subject: Re: [Mpi-forum] Voting results Sent by: mpi-forum-bounces(a)lists.mpi-forum.org On Wed, May 30, 2012 at 7:57 PM, George Bosilca <bosilca(a)eecs.utk.edu> wrote: > In general, the term "simple majority" is biased by our origins. In US it simply means over 1/2 of the valid votes cast (so everything except the absentees). Everywhere else, and this includes some other english speaking countries, a simple majority is the greatest number of parts, in other words the choice that got more votes than any other. Indeed, there are only two countries - none outside of North American - that use the first definition [http://en.wikipedia.org/wiki/Simple_majority] > I can hardly believe I'm the only one with this understanding of the simple majority term. Moreover, I remember live discussion where we clearly stated that there __are__ differences between the "abstain" and the "no"vote, and that one must carefully choose her vote. While these discussions were not in the context of first of second voting, they were at least in the context of readings (which is similarly important as it decide the original fate of a ticket). Maintaining several possible interpretations, and changing the counting before a remote vote doesn't make the process more transparent. I agree completely. If abstain=no then there is no point in having both. > Another valid issue raised during this meeting was about the blessing from the user community. Interestingly enough for a standard targeted toward a user community, such an issue was never brought forward. I think users should be an important factor in our decisions, as important as their contributions to the ongoing discussions. And yes the forum should grant them a significant period of time between the moment when we vote a ticket and the moment when this ticket gets into the standard. Our users will clearly take advantage of such a grace period to thoughtfully evaluate the proposal and raise any issues they might have encountered. We should have done this from day one! No users asked for C++ bindings to be deprecated and a number protested quite vigorously. Many users have asked for active messages and the Forum is has made no progress on this. Even something as simple as MPI_Iwin_create, which was requested by multiple users and had a very valid use case, was shot down in the RMA working group because it was too hard for Brian to implement. The notion that users are a high priority to the Forum is rather farcical as far as I'm concerned. I wish this was not the case. Jeff > On May 31, 2012, at 06:10 , Bronis R. de Supinski wrote: > >> >> All: >> >> Hmm. Quite the controversy. However, the rules as enforced >> in Japan are consistent with my understanding of what they >> have always been. More importantly, they are consistent >> with the wording in the bylaws. Here is what Jeff quoted: >> >> a simple majority is defined as a simple majority >> of those present and eligible to vote. >> >> Those who abstained were present and eligible to vote. >> They did not vote yes. The effect is that they voted >> "no" by this definition. If they did not want their >> vote effectively to be "no" then they should have left >> the room. I recall several instances in which someone >> was out of the room (perhaps even momentarily for a bio >> break) and Jeff recorded their vote as "not present". >> See the definition above -- they then do not count as >> present so they do not figure into the required "yes" count. >> >> As I stated, my understanding of the rules is consistent >> with the interpretation used in Japan. I would object to >> any other interpretation since the by-laws are actually >> clear on this point. I agree that the by-laws should be >> clear in general; while I think they are clear, I would >> not object to a clarifying statement being added to the >> effect that "abstentions are effectively negative votes." >> I think we have many other issues that should be made >> concrete in the by-laws and this is the least important. >> What is required to pass a first reading is probably the >> most obvious issue. >> >> Bronis >> >> >> >> >> On Wed, 30 May 2012, Fab Tillier wrote: >> >>> Jeff Squyres wrote on Wed, 30 May 2012 at 12:12:26 >>> >>>> On May 30, 2012, at 2:57 PM, Jeff Hammond wrote: >>>> >>>>>> The fact that some votes were still recorded as 'abstain' is an indication >>>>>> that this bylaw change was half baked. >>>>> >>>>> Especially when the meeting is attended by so few people due to the >>>>> location. It seems like a weasel tactic to pick a remote location to >>>>> change the by-laws with a single vote. >>>> >>>> >>>> To be clear, the process document states: >>>> >>>> For the purposes of voting, a simple majority is defined as a simple >>>> majority of those present and eligible to vote. >>>> In the context of the document, the phrase "simple majority" is used to >>>> describe what is needed for ballots to pass; this sentence is attempting to >>>> define that phrase. So even though the above sentence looks like a circular >>>> definition, I think it's really an open-ended definition (e.g., a google search >>>> for "simple majority definition" turns up both definitions). >>>> >>>> I was not there and don't know *exactly* what happened, so I'll refrain from >>>> commenting further. >>> >>> If the bylaws are vague, we should clarify them. We should not however reinterpret them at each meeting, and should all agree on a proper interpretation and stick to it, such that ambiguity is removed going forward. Allowing our bylaws to be vague enough to afford a re-interpretation at each meeting does nobody any good. >>> >>> -Fab >>> >>> >>> >>> >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- 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 _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Collection of voting points
by Jeff Squyres 31 May '12

31 May '12

BTW, I want to stress that I think that this is good, healthy debate for the Forum. Thank you all for taking the time to add your voice to the discussion. In reading the thread that continued last night, I wish to make the following points: 1. What is currently being referred to as "the bylaws" is a document that Rich checked into Subversion on May 9, 2011 (i.e., about a year ago). I do recall that there was some discussion about this document before it was posted, but I do not believe that it received the same level of scrutiny as MPI Forum document text, for example. I do not believe that this document had 2 binding votes, either (someone please correct me if I am wrong). *** Meaning: this is a fairly recent document. 2. The first votes I recorded (for MPI-2.2 stuff) was in March, 2008 (see http://meetings.mpi-forum.org/Meeting_details.php and https://svn.mpi-forum.org/svn/mpi-forum-docs/trunk/meetings/2008-03-march/v…) The first MPI-3 vote I see was for NBC in September, 2010. All of these votes were recorded with a "yes votes > no votes" computation. However, as has been pointed out, before the May 2012 meeting, I don't think we've had many (any?) close-enough official votes such that "yes>no" vs. "yes>50%" would have mattered. *** Meaning: there were many votes before the procedures document was published. When paired with the fact that the document was not voted on, referring to that document as "bylaws", I think, is therefore a little misleading. 3. In every single vote record, there are separate columns for yes, no, and abstain (and miss). *** Meaning: This concept of a separate "abstain" result is exactly why I used the original "yes votes > no votes" computation. The existence of "abstain" clearly indicates to me that the Forum intended "abstain" to be different than "no". Note, too, that "abstain" existed in MPI-2 (I wasn't part of MPI-1). 4. Starting in June, 2009, all ballot result web pages contain the following text: NOTE: "Missed" counts the same as "abstain". *** Meaning: This notice did not appear on the 6+1 prior ballot result web pages. I do not remember what happened here, but this *suggests* that there was some discussion in the June 2009 meeting about the exact meaning of "missed". My own belief is that "missed" is just an annotation for inside the forum -- it helps us remember, "Oh, yeah, Joe's plane was late and he missed that vote." Or, "Oh, yeah, Bob must have stepped out of the room for that vote." 5. Regardless of all of the above, there has never been a definition of "present" in the quoted definition of voting: a simple majority is defined as a simple majority of those present and eligible to vote. Does "present" mean "registered to attend the meeting", or "registered to attend the meeting AND physically present in the room at the time of the vote"? *** Meaning: are we saying that there is difference between saying "abstain" and walking out of the room? That seems like a specious distinction, to me. *** Additionally: this specifically calls into question the meaning of the "missed" votes (note that there were no "missed" votes in the Japan meeting, but this point also needs to be clarified). 6. A conspicuous definition that is also absent -- and we have discussed its absence many times -- is that of "quorum". How many voting-eligible organizations are required to constitute critical mass such that a binding vote can take place? Again, as someone correctly pointed out, this was not a factor for the Japan meeting. Here's some stats to show that the Japan votes were representative of the overall Forum: - May 2012 (Japan): 16 orgs voting - March 2012 (Chicago): 17 orgs voting - Jan 2012 (San Jose): 16 orgs voting - ... I didn't bother to go back further *** Meaning: If we're quibbling over the voting rules, the term "quorum" also needs to be defined. === OVERALL SUMMARY: I personally don't have a strong opinion on which set of definitions we end up using. Indeed, Bronis (and Torsten, in IM's with me) make good points about needing a large number of "yes" votes to pass. So this is a good debate to have: if we want to abolish "abstain" (and its synonym, "miss"), then fine -- let's debate and make a conscious, Forum-wide decision to do so. === My objection is that the voting records CLEARLY show that we have treated abstains different than "no"; this is a tradition that extends back to MPI-2. I object to the sudden changing of the ballot rules as I, the secretary, have understood, recorded, and transparently published the ballot results. It seems to be a problem that people were not paying attention to *the last 4 years of ballots*, but that does not mean that the discernment of ballots can change with no discussion by the Forum. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0