POrtable Run-Time System Consortium
Group Meeting: Feb 9, 1995
Pete Beckman, Moderator
Report by Pete Beckman and Suvas Vajracharya
This is the report of the sixth meeting of the PORTS group. The first
meeting took place at a round table (literally) at Supercomputing 93.
This report, as well as other information about the PORTS group can be
http://www.cs.uoregon/paracomp/ports. Pete Beckman moderated the
meeting and Suvas Vajracharya took notes.
Attendees (alphabetically): Pete Beckman (Indiana U)
Nikos Chrisochoides (NPAC, Syracuse U)
Peter Corbeth (IBM Watson)
David DiNucci (Nasa Ames)
Ian Foster (Argonne)
Hubertus Franke (IBM Watson )
Dennis Gannon (Indiana U)
Matt Haines (ICASE)
Paul Havlak (U of Maryland)
Jeff Hollingworth (U of Maryland)
Carl Kesselman (Caltech)
Chuck Koelbel (Rice)
Piyush Mehortra (ICASE)
Bernd Mohr (U of Oregon)
Richard Neves (U of Colorado)
Daniel Quinlan (LANL)
Michel Riviere (IMAG)
Allan Sussman (U of Maryland)
Suvas Vajracharya (U of Colorado)
Hans P. Zima (U of Vienna, ICASE)
There are currently many PORTS0 thread implementations. Go on the PORTS home page to
find the specification, and list of implementations.
Discussion: Extending PORTS0
The group decided that before benchmarks for PORTS0 can be
standardized and shared, PORTS0 must be extended (PORTS0 Version 1.1?)
to include wall-clock timers. A simple vector of timers with
operations such as start, stop, and clear was suggested. A mailing
list of interested people was created.
The group will agree on a timer interface quickly, and mail the
interface to the larger ports group.
Since there are several PORTS0 implementations, the group decided to
develop standard benchmark and validation suites (after timers have
been added to PORTS0). The benchmark suite will measure context
switching, thread creation, etc. The mailing list
was created to discuss and coordinate a standard benchmark code and
methodology. The members will eventually produce a benchmark code and
strict guidelines for collecting data with the benchmark.
The group agreed that some simple validation tests would be a good way
to "certify" that a PORTS0 implementation functioning properly.
However the group decided this was not a high priority, and that the
benchmark suite would make a reasonable validation until someone can
develop a better one.
Future PORTS0 Implementations
Both the Convex Examplar and the Cray T3D were mentioned as parallel
machines that do not yet have a PORTS0 interface.
Discussion: "A PORTS1 Proposal"
The group discussed a proposed extention to PORTS0, entitled A PORTS1
The proposal covered these areas:
Basic machine abstractions
Communication (RPC, MPI, Put/Get):
Remote Procedure Call
There was plenty of discussion. No agreement was reached, but
Timers/Clocks/Events was put at the top of the list of adgenda items,
since it is in the critical path for thread benchmarks.
May 12th (evening) and 13th (all day) in Denver Colorado.
Postscript: Many members of the group went out for snacks and
beer at Bardo's ("The Largest Brewpub in the Western Hemisphere").
The run-time gurus tasted some interesting brews, including an Anaheim
Chili beer, and an incredibly strong stout called Elvis. The grad
students were handed the check, but failed to pay the entire tab.