Base for future multiprocessor standards

[Bob Knighten]

Submitted-by: knighten at pinocchio.encore.com (Bob Knighten)

The POSIX P1003.14 Multi-Processing Working Group is currently
developing a POSIX Standard Profile that will specify base standards for
use by multiprocessors in various contexts.  But there is a problem -
there are hardly any standards that address the needs of
multiprocessors.  Some are being developed.  The proposed standard for
parallel Fortran coming from the ANSI X3H5 committee being one useful
example.  Within the POSIX framework, the MPWG can work with the base
standards groups to provide extensions suitable for multiprocessing, and
we would like to do that.  As a first step we want to locate common
practice in the area that is suitable for standardization within POSIX.
Several areas that seem possible are:

    *  Parallel utilities (e.g. a parallel version of make.)

    *  Resource control (e.g. specifying number, kind of processors a
       program can/must use.)

    *  Thread-safe libraries (signal safe? parallel processing safe?)

    *  Synchronization primitives outside of threads, e.g., between processes.

    *  Specification of a common runtime system, i.e. a common
       multiprocessor underpinning for pthreads, parallel Fortran
       run-times system, parallel Ada run-time systems, and other
       multi-threaded systems.


At the next POSIX meeting (April 15-19 in Chicago) we will consider
whatever examples of current practice we can find.  In preparation I am
trying to gather as much information as possible.

ANY AND ALL COMMENTS ARE WELCOME.

Bob Knighten
Chair, POSIX P1003.14

Encore Computer Corp., 257 Cedar Hill Street, Marlborough, MA 01752
Internet:  knighten at encore.com             (508) 460-0500 ext. 2626
uucp:  {bu-cs,decvax,gould}!encore!knighten

Volume-Number: Volume 23, Number 4
-- 

Sean Eric Fagan, moderator, comp.std.unix.