Front matter text clarifications for the "Collective Communication" Chapter

[tonyskjellum]

Problem

Clarifications are needed now that Persistent Communication has been added.

Proposal

Persistent communication addition has raised the need for clarifying text. This definitely applies to the Collective chapter

Changes to the Text

Chapter 5 (Collective Chapter) -- updated appropriately.

Impact on Implementations

None.

Impact on Users

Clarity for users.

References

Ticket #25 [which has passed 2 votes now] is driving this requirement, but we want to settle this concern separately as agreed in Forum meetings.

Here is the corresponding pull request: mpi-forum/mpi-standard#89
(This is a revised pull requested as of 19-Nov-18)

[tonyskjellum]

mpi32-report-ticket90.pdf

This copy is public.

[tonyskjellum]

A review was done on June 13, 2018 at the MPI meeting in Austin. Updates will be made and a reading will be presented in Barcelona.

[tonyskjellum]

@puribangalore --- Please look at Ticket #80; the latest update adds _X APIs for the persistent collectives; in the process, I've had to go back and enumerate both persistent and persistent _X operations where they have failed to be named throughout Chapter 5. This has got to be related to editorial changes you've been undertaking or plan to do here. In fact, we need to correlate those changes. Ticket #105 will likewise do a review of function enumeration once we add the _X APIs later this week.

[tonyskjellum]

@puribangalore --- we need to update this for the meeting next Thursday ... we need to finish before November 10! We will present in San Jose... do you need help?

[tonyskjellum]

Here's the text we actually want you to read for the two-week deadline review:

mpi32-report-ticket90-19nov18.pdf

[tonyskjellum]

@puribangalore read the chapter; there was various feedback of a general nature.

For instance, issues with commas and capitalization (already addressed to first order).

However, by strawpoll, we voted to reorganize the chapter so that "All Barrier, "All Broadcast," ... "All Exscan." is the order, rather than a parallel section Blocking, Nonblocking, and Persistent. That will necessarily "transpose the 3 sections of operations, and move ahead the discussions, so that the operations will all be "together." This mirror the topologies and other chapters better. This is meant to be part of this ticket... and a careful review is required to be sure we don't change any meanings. That is not supposed to happen just by reordering presentation of the APIs. @puribangalore and @tonyskjellum will do this reordering task and present in Chattanooga in March.

[tonyskjellum]

We did not do the recommended steps for Chattanooga. We are revisiting this task for Chicago.

[tonyskjellum]

Aiming for MPI-4.1 because it is "text tidying." We will wait till all MPI-4.0 work is done before revisiting this.