ARCHITECTURAL DEVELOPMENT
TRACKS
MULTIPLE–PROCESSOR TRACKS
 It can be either a shared-memory multiprocessor or a distributed-memory
multicomputer
Shared-memory track
 Single address space in the entire system
 The track started with C.mmp
NYU Ultrcomputer & Illinois Cedar
 They were developed with a single address space
 Both systems used multistage networks as a system interconnect
 Ultracomputer developed the combining network for fast synchronization among
multiple processors
 NYU developed by Allan Gottlieb in 1983
 The major achievements in Cedar project are in parallel compilers and
performance benchmarking experiments
 Illinois Cedar project was developed by David Kuck in 1987
 Cedar is a cluster-based shared memory multiprocessor
 The system consists of four clusters connected through two unidirectional
interconnection networks to a globally shared memory
Stanford Dash
 NUMA multiprocessor
 Distributed memories forming a global address space
 Cache coherence is enforced with distributed directories
Fujitsu VPP500
 222-processor system with a crossbar interconnect
 Shared memories are distributed to all processor nodes
KSR 1
 It’s a typical COMA model
 Kendall Square Research,1990
IBM RP3 & BBN Butterfly
 These are two large-scale multiprocessors
 Both use multistage networks but with different interstage connections
 RP3, the Research Parallel Processing Prototype, was a research vehicle for
exploring the hardware and software aspects of highly parallel computation.
 RP3 was a shared-memory machine that was designed to be scalable to 512
processors; a 64-processor machine was in operation from October 1988
through March 1991
 The BBN Butterfly was a massively parallel computer built by Bolt, Beranek
and Newman in the 1980s.
 Each machine had up to 512 CPUs, each with local memory, which could be
connected to allow every CPU access to every other CPU's memory, although
with a substantially greater latency (roughly 15:1) than for its own. The CPUs
were commodity microprocessors.
 The memory address space was shared.
Message-Passing Track
Cosmic Cube
 The Caltech Cosmic Cube was a parallel computer, developed by Charles
Seitz and Geoffrey from 1981 onward
 It leads the development of message-passing multicomputers
 It was an early attempt to capitalise on VLSI to speed up scientific
calculations at a reasonable cost
 Intel has produced a series of medium –grain hypercube computers(the iPSCs)
 The nCUBE 2 is also a hypercube configuration
 Paragon is the latest Intel system
 Mosaic and the MIT J-Machine are on the research track which are the two fine-grain
multicomputers
MULTIVECTOR & SIMD TRACKS
Multivector track
 These are traditional vector supercomputers
 The CDC 7600 was the first vector dual-processor system
 The Cray and Japanese supercomputers followed the register-to-register
architecture
 Cray 1 leads the multivector development in 1978
 The latest Cray/MPP is a massively parallel with distributed shared memory
 Its work as a back-end accelerator with Cray Y-MP Series
The SIMD Track
 The Illiac IV leads the construction of SIMD computers
 The Goodyear MPP, the AMT/DAP610, and the TMC/CM-2,are all SIMD
machines built with bit-slice PEs
 The CM-5 is a synchronized MIMD machine executing in a multiple-SIMD mode
 The word-wide PEs is used in :
 BSP was a shared-memory SIMD machine built with 16 processors
updating a group of 17 memory modules synchronously
 The GF11 was developed at the IBM Watson Laboratory for scientific
simulation research use
 The MasPar MP1 is the only medium-grain SIMD computer currently in
production use
Architectural Development Tracks
Architectural Development Tracks

Architectural Development Tracks

  • 1.
  • 2.
    MULTIPLE–PROCESSOR TRACKS  Itcan be either a shared-memory multiprocessor or a distributed-memory multicomputer Shared-memory track  Single address space in the entire system  The track started with C.mmp
  • 3.
    NYU Ultrcomputer &Illinois Cedar  They were developed with a single address space  Both systems used multistage networks as a system interconnect  Ultracomputer developed the combining network for fast synchronization among multiple processors  NYU developed by Allan Gottlieb in 1983
  • 4.
     The majorachievements in Cedar project are in parallel compilers and performance benchmarking experiments  Illinois Cedar project was developed by David Kuck in 1987  Cedar is a cluster-based shared memory multiprocessor  The system consists of four clusters connected through two unidirectional interconnection networks to a globally shared memory
  • 5.
    Stanford Dash  NUMAmultiprocessor  Distributed memories forming a global address space  Cache coherence is enforced with distributed directories
  • 6.
    Fujitsu VPP500  222-processorsystem with a crossbar interconnect  Shared memories are distributed to all processor nodes KSR 1  It’s a typical COMA model  Kendall Square Research,1990
  • 7.
    IBM RP3 &BBN Butterfly  These are two large-scale multiprocessors  Both use multistage networks but with different interstage connections  RP3, the Research Parallel Processing Prototype, was a research vehicle for exploring the hardware and software aspects of highly parallel computation.  RP3 was a shared-memory machine that was designed to be scalable to 512 processors; a 64-processor machine was in operation from October 1988 through March 1991
  • 8.
     The BBNButterfly was a massively parallel computer built by Bolt, Beranek and Newman in the 1980s.  Each machine had up to 512 CPUs, each with local memory, which could be connected to allow every CPU access to every other CPU's memory, although with a substantially greater latency (roughly 15:1) than for its own. The CPUs were commodity microprocessors.  The memory address space was shared.
  • 9.
  • 10.
    Cosmic Cube  TheCaltech Cosmic Cube was a parallel computer, developed by Charles Seitz and Geoffrey from 1981 onward  It leads the development of message-passing multicomputers  It was an early attempt to capitalise on VLSI to speed up scientific calculations at a reasonable cost
  • 11.
     Intel hasproduced a series of medium –grain hypercube computers(the iPSCs)  The nCUBE 2 is also a hypercube configuration  Paragon is the latest Intel system  Mosaic and the MIT J-Machine are on the research track which are the two fine-grain multicomputers
  • 13.
    MULTIVECTOR & SIMDTRACKS Multivector track  These are traditional vector supercomputers  The CDC 7600 was the first vector dual-processor system  The Cray and Japanese supercomputers followed the register-to-register architecture  Cray 1 leads the multivector development in 1978  The latest Cray/MPP is a massively parallel with distributed shared memory  Its work as a back-end accelerator with Cray Y-MP Series
  • 16.
    The SIMD Track The Illiac IV leads the construction of SIMD computers  The Goodyear MPP, the AMT/DAP610, and the TMC/CM-2,are all SIMD machines built with bit-slice PEs  The CM-5 is a synchronized MIMD machine executing in a multiple-SIMD mode
  • 17.
     The word-widePEs is used in :  BSP was a shared-memory SIMD machine built with 16 processors updating a group of 17 memory modules synchronously  The GF11 was developed at the IBM Watson Laboratory for scientific simulation research use  The MasPar MP1 is the only medium-grain SIMD computer currently in production use