[Mpi-forum] MPI-next

Rolf Rabenseifner rabenseifner at hlrs.de
Sat Feb 22 00:53:51 CST 2014


Dear all,

as input for your discussions I'll give you some background about the 
MPI-2.1/2.2/3.0 book selling which includes some lessens learned,
which may influence your decisions on MPI-3.1 and 4.0,
see conclusions at the end.

Facts and implications:

- Oct.2012 - Feb.2014 we sold 2238 MPI books which is 33 books/month.

  ==> Currently remaining 455 MPI-3.0 books will be sold out
      about end of March 2015


- It is no problem to have 100 unsold books at the end.

- Print-on-demand does not work for the size of the MPI book.
- Normal printing via a publisher does not work for that low price.
 
- We have to produce about 1000 books to keep the price such low,
  i.e., 25$ plus shipping or 19.50 Euro plus shipping

  ==> Selling time should be about 30 month.


- About 30% of MPI programmers who physically get in touch with the book,
  buy the book (it is the cheapest useful screen extension they can buy ;-)

  ==> I expect a similar rate in the MPI Forum

  ==> Therefore, I expect that the 70% non-MPI-book-buying majority
      in the forum is mainly deciding for the 30% MPI-book-buying 
      minority. I hope that this majority keeps this in mind and
      makes good decisions for the minority. 


Conclusions:
------------

The second option
> - MPI 3.1 for minor additions and incl. errata, 
>   MPI 4.0 for Major additions
together with
> In particular, we discussed shooting for making the March
> meeting the last one during which additions for MPI 3.1 could be
> read, which would get us to a standard by December.
seems to be really good.

It implies the printed book at end of Jan. 2015.

The major question is the time-line for MPI-4.0
which should be from the book publisher's 
view-point June 2017 (about 30 month after 3.1).

Reality showed the total MPI-3.0 process needed 58 month,
from Dec.2007 - Sep.2012.

The MPI-4.0 project started already within MPI-3.0,
>From MPI-3.0 Sep. 2012 until projected MPI-4.0 June 2017 
will be again 58 month, which seems a healthy plan.
New standards need not only to be implemented,
the implementations also need to be optimized,
and both needs time.

Previous and next colors for MPI books in the rainbow:
2.1 yellow
2.2 orange (yellow and orange were in wrong sequence)
3.0 green
3.1 perhaps cyan
4.0 perhaps blue
(5.0 purple)

This was more than only my 5 cents.

Best regards
Rolf  


PS: Total statistics
- MPI-2.1  916 printed, 738 sold Jul.08 - Sep.09 (14 month) = 53 books/month
- MPI-2.2  921 printed, 900 sold Oct.09 - Sep.12 (36 month) = 25 books/month
- MPI-3.0 1055 printed, 600 sold Oct.12 - Feb.13 (17 month) = 35 books/month
          
    

----- Original Message -----
> From: "Martin Schulz" <schulzm at llnl.gov>
> To: "Main MPI Forum mailing list" <mpi-forum at lists.mpi-forum.org>
> Sent: Tuesday, December 17, 2013 5:02:07 AM
> Subject: [Mpi-forum] MPI-next
> 
> Hi all,
> 
> During the last MPI forum meeting in Chicago we had a discussion on
> what MPI-next should be and by when we should have it out. We
> basically looked at five options:
> 
> - Skip any MPI 3.x and go straight to MPI 4.0
> - MPI 3.1 for minor additions and incl. errata, MPI 4.0 for major
> additions
> - MPI 3.1 only for errata, MPI 4.0 for anything else
> - MPI 3.1 for errata, MPI 3.2 for small enhancements, then MPI 4.0
> - MPI 3.0.1 for errata, MPI 3.1 for small enhancements, then MPI 4.0
> 
> Out of those, the majority favored the second option - create an MPI
> 3.1 that includes errata and minor additions, followed by an MPI 4.0
> later. We left what is minor and what is not out of the discussion
> for now, though, with the idea to look at the tickets that are ready
> one by one. Further, a majority favored a rather short timeline for
> MPI 3.1. In particular, we discussed shooting for making the March
> meeting the last one during which additions for MPI 3.1 could be
> read, which would get us to a standard by December.
> 
> If someone feels this is too short of a timeline and would cause
> important things for MPI 3.1 to not be included, please speak up and
> we can continue the discussion over email.
> 
> Thanks,
> 
> Martin
> 
> 
> 
> 
> ________________________________________________________________________
> Martin Schulz, schulzm at llnl.gov, http://people.llnl.gov/schulzm
> CASC @ Lawrence Livermore National Laboratory, Livermore, USA
> 
> 
> 
> _______________________________________________
> mpi-forum mailing list
> mpi-forum at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum
> 

-- 
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner at hlrs.de
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: Room 1.307)



More information about the mpi-forum mailing list