john@uunet.UU.NET (John C. Sieg) (05/03/91)
COLLOQUIUM ANNOUNCEMENT
Computer Science Department
College of Arts and Sciences, University of Lowell
DATE: Wednesday, May 8, 1991
TIME: 3:00 p.m., refreshments at 2:30
PLACE: University of Lowell
Wannalancit 116 (175 Cabot St., Lowell, MA)
(directions available from john@cs.ulowell.edu)
SPEAKER: Linda Wilkens and James Canning
Department of Computer Science
University of Lowell
Lowell, MA 01854
TITLE: Using Extended Petri Nets to Map Parallel Algorithms
onto Parallel Architectures
ABSTRACT:
The objective of this research is to develop an
automated methodology which will efficiently map parallel
algorithms onto parallel architectures. Our proposed itera-
tive technique cycles through three processes: model, exe-
cute, and map, to converge on an efficient solution. The
model process relies on extended Petri nets to individually
model a parallel algorithm and a parallel architecture.
These two models are then fused to represent the algorithm
loaded into the architecture. The fused models are subse-
quently executed, providing feedback to the mapper process.
The mapper uses this information to converge on a more effi-
cient mapping. Our methodology can also be used to design
parallel architectures which will efficiently execute a
given parallel algorithm. We are implementing the methodol-
ogy and verifying its effectiveness, using the ULowell
dataflow coprocessor as a testbed. This talk will describe
the modeling process and show how we model a fine-grain
parallel machine. The results of this work can be applied to
a wide range of parallel computing models, making it much
easier for programmers to maximize the power of their paral-
lel machines.
--
=========================== MODERATOR ==============================
Steve Stevenson {steve,fpst}@hubcap.clemson.edu
Department of Computer Science, comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell