CALL FOR PAPERS

			3rd Workshop on
	Job Scheduling Strategies for Parallel Processing

	   in conjunction with IPPS '97, April 5, 1997
	    University of Geneva, Geneva, Switzerland


As large parallel computers become more popular, scheduling strategies
become more important as a means of balancing the need for exclusive
use of the machine's resources and the desire to make these resources
readily available to many diverse users.  Neither sign-up sheets,
naive time-slicing, nor naive space-slicing are suitable solutions.
Moreover, there appears to be a divergence between what is studied,
modeled, and analyzed in academic circles and the actual, sometimes
ad-hoc, scheduling schemes developed by vendors and large
installations.

Continuing the tradition established at IPPS'95,
the workshop is intended to attract people from academia, supercomputing
centers, national laboratories, and parallel computer vendors to
address resource management issues in multiuser parallel systems, and
attempt to resolve the conflicting goals such as short response times
for interactive work, minimal interference with batch jobs, fairness
to all users, and high system utilization.  We hope to achieve a
balance between reports of current practices in large and heavily-used
installations, proposals of novel schemes that have not yet been
tested in a real environment, and realistic models and analysis.
The emphasis will be on practical designs in the context of real
parallel operating systems.

Topics of interest include:
<> Experience with scheduling policies on current systems
<> Performance implications of scheduling strategies
<> Fairness, priorities, and accounting issues
<> Workload characterization and classification
<> Support for different classes of jobs (e.g. interactive vs. batch)
<> Static vs. dynamic partitioning
<> Time slicing and gang scheduling
<> Interaction of Scheduling with the model of computation
<> Interaction of scheduling with memory management and I/O
<> Load estimation and load balancing
<> Scheduling on heterogeneous nodes (e.g. with different amounts of memory)
<> Performance metrics to compare scheduling schemes

REGISTRATION:

The workshop is open to all participants of IPPS '97.
For general IPPS '97 information, see URL http://cuiwww.unige.ch/~ipps97.

ORGANIZERS:

Dror Feitelson, Hebrew University (feit@cs.huji.ac.il)
Larry Rudolph, Hebrew University & MIT (rudolph@theory.lcs.mit.edu)

PROGRAM COMMITTEE:

Nawaf Bitar, Silicon Graphics
David Black, OSF
Atsushi Hori, RWCP
Mal Kalos, Cornell Theory Center
Richard Lagerstrom, Cray Research
Miron Livny, University of Wisconsin
Virginia Lo, University of Oregon
Reagan Moore, SDSC
Bill Nitzberg, NASA Ames
Uwe Schwiegelshohn, University Dortmund
Ken Sevcik, University of Toronto
Mark Squillante, IBM Research
John Zahorjan, University of Washington

SUBMISSIONS:

Papers should be no longer than 20 pages, including figures and
references.  All papers will be reviewed, and a proceedings will be
distributed at the workshop (in previous years, a post-workshop
proceedings was also published by Springer-Verlag in the Lecture Notes
on Computer Science series).
Send uuencoded postscript of the paper by e-mail to:

   feit@cs.huji.ac.il

or 6 hard copies of the paper to:

   Dror Feitelson
   Institute of Computer Science
   Givat Ram Campus, Ross 28
   The Hebrew University
   91904 Jerusalem, ISRAEL

Please include name, address, phone, and email of contact author.

DEADLINE:       December 27, 1996
NOTIFICATION:   February 7, 1997
FINAL COPY DUE: March 7, 1997