The Queen's University of Belfast

Parallel Computer Centre
[Next] [Previous] [Top]
Classification
Flynn's Taxonomy
- Machines classified by instruction delivery mechanism and data stream
- A four character code is used
- first 2 define instruction stream type
- second 2 define data stream type
- Streams may be either single or multiple
- The possible variations are
Examples
- SISD: traditional Von Neuman architecture
- SIMD: original massively parallel machines
- AMT DAP
- MASPAR
- Connection Machine
- MISD: data flow machine - experimental
- MIMD: flexible multi-processor of which there are 2 sub-classes
- MIMD SM - Shared Memory
- MIMD DM - Distributed Memory
SISD
Simple Diagrammatic Representation

SIMD
Simple Diagrammatic Representation

MIMD SM
Shared Memory
- All processors share the same main memory
- Relatively simple to program
- Direct use of multi-tasking techniques such as semaphores and Mutex
- All of the memory is accessible by every processor - small sections may be temporarily locked by one processor
- Examples
- CRAY YMP, CRAY C90
- IBM 3090
- VAX 9000
- Sequent Symmetry
Simple Diagrammatic Representation

MIMD DM
Distributed Memory
- Memory is distributed among the processors ie. each processor has its own memory - alternative acronym NUMA (Non Uniform Memory Access)
- Connectivity between processors is required
- Bus or switch: potential bottleneck
- all-to-all connections (increases with n2)
- specialized topology (hard wired or using routing/switching processors)
- networked workstations - usually a broadcast/multi-access medium such as
- Examples
- Transputer based: Parsys, Meiko
- DEC Alpha: Cray T3D
- i860: Paragon, Delta, iPSC
- SGI workstation cluster
Simple Diagrammatic Representation

Flynn's Taxonomy
An Analogy
- Consider the check-in desks at an airport
- SISD: a single desk
- SIMD: many desks and a supervisor with a megaphone issuing instructions which every desk obey
- MIMD: many desks working at their own pace, synchronising via a central database
[Next] [Previous] [Top]
All documents are the responsibility of, and copyright, © their authors and do not represent the views of The Parallel Computer Centre, nor of The Queen's University of Belfast.
Maintained by Alan Rea, email A.Rea@qub.ac.uk
Generated with CERN WebMaker