Vortrag: Dr. Thomas Sterling, 12.4.1999

[Maria Cherry]

                              UNIVERSITAET WIEN 
              INSTITUT FUER SOFTWARETECHNIK UND PARALLELE SYSTEME
                                 gemeinsam mit 
                                    VCPC 
               EUROPEAN CENTRE FOR PARALLEL COMPUTING AT VIENNA 
                       
                        FWF-Projekt SFB F011 "AURORA"


        EINLADUNG ZU EINEM VORTRAG IM RAHMEN DES AURORA-KOLLOQUIUMS
                         
            
                                               
                   Future Directions in High End Computing

			  
		  	    Dr. Thomas Sterling

		   Center for Advanced Computing Research
		      California Institute of Technology
		  	 	   and
		     High Performance Computing Group
		      NASA Jet Propulsion Laboratory


                  
                  ZEIT: Montag, 12. 4. 1999, 17.15 Uhr s.t.
          ORT: Institut fuer Softwaretechnik und Parallele Systeme
                   1090 Wien, Liechtensteinstrasse 22, 
                          Seminarraum, Mezzanin



Abstract:

High end computing system architecture is dominated by Massively
Parallel Processing (MPP) systems with major systems provided by
manufacturers in the US and Japan. However, new opportunities in
technology and architecture as well as challenges encountered in MPP
programmability and scalability are driving innovative directions for
future system structure in the next five years. Several important trends
will be discussed in this presentation that are likely to be reflected
in high end parallel computer systems. Beowulf-class systems exploit
mass market commodity off-the-shelf Personal Computers (PC) in clusters
integrated by conventional local area network (LAN) technology to
provide order of magnitude improvement in price-performance with respect
to commercial MPPs for those application algorithms suitable for loosely
coupled parallel systems. Processor-in-Memory (PIM) technology merges
logic with memory cells to provide X100 memory bandwidth and X10 power
to performance advantage over conventional processor/memory
configurations. Advanced technologies are maturing and will provide X100
throughput over semiconductor based methods. These include
superconductor RSFQ logic and optical wave-division multiplexing
communications. Multithreaded architecture and advanced proactive
adaptive resource management techniques are providing a powerful
approach to overcoming performance degradation due to latency and
overhead. This seminar will present recent research results from several
research projects that are leading future directions in high end
computer architecture. Beowulf-class computing, the DIVA PIM
architecture project, and the Hybrid Technology MultiThreaded (HTMT)
architecture program will all be presented. Dr. Sterling is recognized
as the father of Beowulf-class clustered computing using Linux and low
cost PCs. He has led the architecture work on the DIVA PIM project, and
is the Principal Investigator of the HTMT project that is developing a
Petaflops scale parallel computer using Superconductor logic and optical
communications. Dr. Sterling will also suggest long term trends as
represented in a new family of parallel structures, Continuum Computer
Architecture.