taylor@hplabsc.UUCP (07/03/86)
This article is from Nathaniel Mishkin <apollo!mishkin>
and was received on Thu Jul 3 00:34:33 1986
CALL FOR PARTICIPATION
IEEE Workshop on Large Grained Parallelism
October 27-29, 1986
Biltmore Plaza Hotel
Providence, Rhode Island
Large networks of personal workstations are becoming common, and are
likely to become both larger and more common. Other loosely-coupled
multi-computers, such as the CalTech Cosmic Cube, are also starting to
become available. The aggregate computing power of such systems could
easily exceed that of the fastest supercomputer. Harnessing such power
has, however, proved elusive. Much work on algorithms that can exploit
parallelism has been done, particularly in the context of vector/array
processors. However, the underlying tools often used to implement such
algorithms are not right for networks of workstations. Such language
constructs as `cobegin/coend', communicating sequential processes, or
processes and monitors running in a single address space, focus
attention on situations where creating parallel threads of execution
and communications between them are inexpensive. I.e., they are
oriented toward "fine-grained parallelism". Unfortunately, in the
loosely-coupled multi-computer context, the costs of creating and
communicating among concurrent threads on different nodes is often too
high for such techniques to be useful.
So, what we have is a solution in search of some good problems --
i.e., a powerful network of computers which are nearly idle most of
the time, just waiting for something useful to do. What kind of
problems can be decomposed into a large number of pieces which are
nearly independent, which compute for a long time relative to the cost
of creating them, and which do not need much communication between
themselves? What algorithms, in other words, can use "large-grained"
parallelism?
Possible Issues:
- What kind of computations can profitably use a network of
workstations or other loosely-coupled multi-computers?
- What tools are needed to make it easier to write and execute such
applications?
- How is the autonomy of the individual and the workstation
preserved?
- What metrics is appropriate for judging the performance of such a
system?
- What kind of language support might be appropriate for specifying
such computations?
- What other problems are related to this one?
To facilitate the dialog that is so valuable in a workshop setting,
attendance will be limited to approximately 40 individuals. Potential
participants should prepare a statement of three pages or less
describing specific related experiences and the participant's
interests and future directions related to the workshop topic.
Submissions may be via electronic mail to the program chair or general
chair (adresses below), or send five copies via US Mail to:
Paul J. Leach
Apollo Computer Inc.
330 Billerica Rd.
Chelmsford, MA 01824
Submission deadline: July 21, 1986.
Acceptance notification: August 15, 1986.
Registration Deadline: September 15, 1986.
Workshop Organizing Committee:
General Chair: John Stankovic (UMass Amherst; stankovic@umass-ece.csnet)
Program Chair: Paul J. Leach (Apollo Computer; ...wanginst!apollo!pjl)
Local Arrangements: Tom Doeppner (Brown University)
Program Committee: Guy Almes (Rice University)
Lee Hollaar (University of Utah)
Luis F. Cabrera (IBM Almaden)
Costs:
IEEE Members: $100 (late registration $125)
Non-members: $125 (late registration $160)