🚀 go-pugleaf

                    Center for High Performance Computing
                                Colloquium

	    Performance Environments for Scalable Parallel Systems

			     Daniel A. Reed

		     Department of Computer Science
		         University of Illinois
		         Urbana, Illinois  61801

   The complexity of parallel systems makes a priori performance
   prediction difficult and experimental performance analysis crucial.  A
   performance data analysis and data presentation environment is needed
   to access the effects of architectural and system software variations.
   We describe the Pablo performance analysis environment, a software
   toolkit whose design is based on the lessons learned from two previous
   generations of performance data analysis software.  Pablo contains a
   set of performance data transformation modules that can be
   interconnected in user-specified ways.  It is the responsibility of the
   environment infrastructure to hide details of module interconnection
   and data sharing.  The environment is written in C++ with the graphical
   displays based on X windows and the Motif toolkit.  It allows users to
   interconnect and configure modules graphically to form an acyclic,
   directed data analysis graph.  Performance trace data are represented
   in a self-documenting stream format that includes internal definitions
   of data types, sizes, and names.  This self-documenting data format is
   the lingua franca for module data sharing.  Because experience has
   shown that scientific data visualization is experimental, rarely can
   the most appropriate visualization idiom be predicted in advance.  In
   addition, psychological studies have shown that some data are more
   quickly grasped when heard rather than seen.  Thus, Pablo supports the
   use of sonic data presentation in addition to the traditional use of
   visual techniques, and we have just begun to explore the applicability
   of virtual reality techniques to performance data analysis and
   presentation for massively parallel systems.  Although the environment
   is now operational, much work remains to be completed, including
   extensions of the environment infrastructure to support new data
   analysis modules, graphical displays, and parallel execution.

   Using the Pablo environment as basis, we also sketch our plans for
   resource management studies on massively parallel systems.  Though the
   adoption of Unix or Mach as a standard operating system for parallel systems
   promises to greatly improve the system software substrate, the
   facilities for data abstraction, dynamic distribution of load, and
   high-performance input/output are rudimentary at best and more often
   non-existent.  Moreover, many basic questions about traditional
   resource management remain unanswered in the context of large scale,
   distributed memory parallel systems.
------------------------------------------------------------------------------
 Date:  Thursday, November 12, 1992
 Time:  10:30 AM
Place:  Center for High Performance Computing 
	Worcester Polytechnic Institute
	293 Boston Post Road West,  Marlborough, MA
 Info:  Contact Rick LaRowe @ (508) 624-7400 x610 or rlarowe@chpc.org
------------------------------------------------------------------------------