mrb@sei.cmu.edu (Mario Barbacci) (10/27/87)
[ Thanks, Mario. --DL ]
Summary of the Second Workshop on Large-Grained Parallelism
The Second Workshop on Large-Grained Parallelism took place October 11-14,
1987, in Hidden Valley, Pennsylvania. The workshop was organized by the
Software Engineering Institute and the Department of Computer Science,
Carnegie Mellon University, with the cooperation of the IEEE Computer
Society.
The purpose of the workshop was to bring together people whose interests lie
in the areas of operating systems, programming languages, and formal
models for parallel and distributed computing. The emphasis of the
workshop was on large-grained parallelism or parallelism between concurrent
programs running on networks of possibly heterogeneous computers rather
than parallelism within a single process or thread of control. Aspects of
large-grained parallelism that were common to most participants'
interests were fault-tolerance, heterogeneity, and real-time
applications.
Ninety abstracts were submitted for review by the program committee and
the authors of forty of these abstracts were sent acceptance
letters and invitations to attend the workshop. To provide more time for
discussion and audience participation, only fifteen authors were asked
to give twenty-five minute talks based on their abstracts. The rest
of the abstracts were summarized by discussion leaders. The workshop
was divided into five sessions of talks and two parallel sessions of
discussion. The five general areas covered by the talks were:
scheduling, distributed languages, real-time languages and models,
operating system support, and applications. There were parallel
discussions on scheduling and distributed languages, and on real-time and
operating system support.
There was a reasonable balance among the participants with regard to
efficiency concerns on the one hand, e.g., by the software and
hardware systems and application builders, and correctness concerns on the
other, e.g., by the real-time modelers and language designers. We
identified a number of key challenges:
- Distributed systems, languages, environments
* Make transactions efficient. Integrate them into the operating
system.
* Implement applications that demonstrate how to use transactions
at both the programming language and operating system levels.
* Identify applications other than databases to motivate the need
for multi-site transaction-based systems.
- Real-time systems, models, scheduling
* Devise and test analytical models for distributed scheduling of
tasks that range in degrees of computational complexity.
* Show the correspondence between physical time and logical time
using a formal modeling approach.
* Identify a set of programming and specification language
primitives that capture and abstract from real-time events of
interest.
In the year that elapsed since the first workshop on large-grained
parallelism that took place in Providence, Rhode Island, a number of the
issues related to large-grained parallelism became more focused, as
evidenced by the topics and the quality of the abstracts submitted.
Considering the wide range of interests and background of the
participants, the success of this workshop is a good omen for future
meetings.
The workshop proceedings will be published as an SEI technical report. For a
copy of the report contact:
Documentation Services
Software Engineering Institute
Carnegie Mellon University
Pittsburgh, Pensylvania 15213
or, by E-Mail on the ArpaNet:
barbara.zayas@sei.cmu.edu
------------------------------------------------------------------------
Mario R. Barbacci, (ArpaNet:barbacci@sei.cmu.edu)
Software Engineering Institute, CMU, Pittsburgh PA 15213, (412) 268-7704