TEMPLE UNIVERSITY SYNERGY PROJECT
CENTER FOR ADVANCED COMPUTING AND COMMUNICATIONS
Welcome:
// Local time:
WHAT IS SYNERGY?
- SYNERGY:
syn-er-gy \'sin-er-je^-\
[NL synergia, fr. Gk synergos working together]
:SYNERGISM;
broadly: combined action or operation.
- The SYNERGY system [U.S. Patent #5,381,534]
was designed as an effective parallel programming platform using
distributed computers. It supports automatic processor
assignments, automatic load balancing and
automatic SIMD worker fault detection and recovery .
SYNERGY Version 3.0
has been ported to the following platforms:
SUN4 (SunOS), IBM RS6K (AIX), Silicon Graphics (IRIX),
HP (HP/UX), DEC PMAX (ULTRIX), DEC alpha (OSF/1).
WHY SYNERGY?
- SEPARATION of functional programming from process coordination and
resource management functions
-- Relieving programmers from the chores of process coordination
and resource management improves programming productivity.
It also improves performance since hard-wired controls are
no longer in the programs. Runtime optimization becomes
possible.
- UNIFIED "shared memory" and "distributed memory" parallel programming
styles
-- This point is probably for re-engineering existing parallel codes.
- AUTOMATIC TASK SCHEDULING
-- POFP is a dataflow method. Processes can be automatically scheduled
using natural application dependent dataflows.
- AUTOMATIC PROCESSOR ASSIGNMENTS
-- Using POFP, a compute intense program is decomposed into a network
of interconnected non-compute intense programs. Automatic
processor assignment ensures that the potential parallelism
can be realized at runtime.
- AUTOMATIC LOAD BALANCING
-- For POFP programs, load balancing is only needed among SIMD
workers. Using a modified "dollar cost averaging" method, one
can effectively control the synchronization overhead in
fluctuating environments.
- AUTOMATIC FAULT DETECTION AND RECOVERY
-- Synergy V3.0 also provides automatic detection and recovery of
compute intense SIMD workers. Using POFP, this fault tolerance
facility has very little overhead. Further, it requires no
special programming effort.
- DISCIPLINED PARALLEL PROCESSING AND NEW ARCHITECTURE
-- We have developed a quantitative discipline for composing parallel
programs
( Evaluation of Parallel Programs (Postscript, 240k) ). For a tutorial of the Synergy system see
( Synergy Tutorial (Postscript, 171k) ).
HOW TO GET SYNERGY?
- Site-license fees:
- $2,500 (with ASCII interface)
- $3,000 (with Motif GUI)
- Contact Dr. Antonio Goncalves, Vice Provost, Room 406, USB, Temple University, Philadelphia, PA 19122. Phone: (215) 204-7662, Fax: (215) 204-7486.
- The public domain evaluation package (no processor limit)
- Purchasing source code License -- for more information, contact Dr. Yuan Shi
(shi@falcon.ocis.temple.edu)
SYNERGY ON-LINE
REFERENCES
Synergy Team Members:
ADMINISTRATOR
OTHERS
Yuan Shi
Last Update: August 10, 1995.
