(j3.2006) (SC22WG5.4924) WG5 ballot on first draft TS 18508, Additional Parallel Features in Fortran

John Reid John.Reid
Fri Mar 8 07:06:04 EST 2013


In the strategic plan for the coarray TS agreed at Markham (N1925) there 

    Draft TS constructed by J3       2013-02
    WG5 ballot on draft TS           2013-03

I am pleased to say that we are on schedule.

The draft TS is N1967 and the ballot is N1968. Both are attached. There 
are background notes in N1968 that you should read before reading N1967 

The ballot ends on 8 April.

Best wishes,


-------------- next part --------------
                                        ISO/IEC JTC1/SC22/WG5 N1968
                       WG5 letter ballot on N1967

                        John Reid, 8 March 2013

This is a WG5 letter ballot on N1967, the first draft DTS for TS 18508, 
Additional Parallel Features in Fortran.

N1967 has the same content as J3/13-251. It was prepared by J3 from 
J3/12-201 (collectives and atomic operations), 13-231r5 (teams), and 
13-232r3 (events). The basic requirements are specified in WG5/N1930. 
Based on technical discussions within J3 during preparation of 13-251, 
three enhancements to the original requirements were incorporated.

1) The basic team model is very useful for the uses outlined in N1930,
and that functionality is included. However, there are other use cases
that need only a fast synchronization of a subset of the images. With
the infrastructure for teams already established, addition of a SYNC
TEAM statement was possible with little extra text or implementation
work.  This was added in 13-231r5.

2) The general event model in N1930 addresses the functionality
required, but does not provide for optimal performance for the widely
used case of waiting on a local event. A second event type,
LOCAL_EVENT_TYPE, was added by 13-232r3 to enable this restricted, but
better performing, capability.

3) Resiliency of parallel programs to partial system failures has
become an increasingly important issue, particularly for systems with
heavy Fortran usage of the size that will be available in the time
frame when implementations of this TS are expected. Addressing this
issue is a main feature of the upcoming MPI 4 spec.  The traditional
Fortran model cannot offer a viable solution to this problem because
of the underlying assumption that the number of images is fixed for
the entire program execution. The introduction of teams provides a
mechanism that allows for the effective number of images to
change. With this capability, it is possible to have a framework for
addressing the problem of images that fail during execution of a
program.  As part of the team feature in 13-231r5 minimal resiliency
capabilities were added that can notify the program if an image has
failed and provide a list of which images have failed. With this
information and the ability to form a new team that excludes the
failed images, the programs implementing fault-tolerant algorithms
might be able to profitably continue execution.

Please answer the following question "Is N1967 ready for forwarding to 
SC22 as the DTS?" in one of these ways. 

1) Yes.
2) Yes, but I recommend the following changes. 
3) No, for the following reasons.
4) Abstain.

This is an individual vote. Please send your vote to sc22wg5 at open-std.org 
to arrive by 9 a.m. (UK time) on 8 April 2013. 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: N1967.pdf
Type: application/pdf
Size: 229443 bytes
Desc: not available
Url : http://mailman.j3-fortran.org/pipermail/j3/attachments/20130308/41cb6ef3/attachment-0001.pdf 

More information about the J3 mailing list