Conceptual approach to general purpose parallel computer architecture
Thesis/Dissertation
·
OSTI ID:5689602
One approach to speedup the solution of problems by computers is to use concurrent execution. Parallel computers may be viewed as a set of sequential processors linked by an interconnection network. Networks suitable for one algorithm may not be suitable for another. This thesis is concerned with an architecture for a parallel computer that is suitable for the broadest possible class of algorithms. This architecture is called GEM, for generalized machine. GEM provides performance commensurate with any other parallel computer provided that GEM has at least as many processors, GEM has the same or larger instruction set, and the sequential processors of GEM are at least as fast. The architecture of GEM is theoretical, not practical. That is, it is concerned only with logical limitations, not with physical or economic limitations. However, the GEM theory provides the basis for the CHoPP (Columbia Homogeneous Parallel Processor) design, which is a practical implementation of the theory. GEM contains a memory that is shared equally by the processors via an interconnection network. This network is a packet switching network. A conflict filter function has been devised which avoids network and memory congestion for multiple simultaneous references to any one memory location. Memory addresses are assigned to memory banks by a hash function to avoid network and memory congestion for simultaneous references to different memory locations. The user views GEM as containing an unlimited number of processors, called virtual processors. Each real processor multiplexes among a set of virtual processors.
- OSTI ID:
- 5689602
- Country of Publication:
- United States
- Language:
- English
Similar Records
Multicluster approach to computer vision
Bipartite memory network architectures for parallel processing
Parallel linear solvers for reservoir simulation: A generic approach for existing and emerging computer architectures
Conference
·
Sat Oct 01 00:00:00 EDT 1983
·
OSTI ID:5363640
Bipartite memory network architectures for parallel processing
Book
·
Sun Dec 31 23:00:00 EST 1989
·
OSTI ID:5123927
Parallel linear solvers for reservoir simulation: A generic approach for existing and emerging computer architectures
Journal Article
·
Mon Jan 31 23:00:00 EST 1994
· SPE Computer Applications (Society of Petroleum Engineers); (United States)
·
OSTI ID:7115894