Hypercomputers, algorithmic structures and hyperware: A reconfigurable, massively parallel architecture
Reconfigurable, parallel architectures offer much greater promise for high performance over a diverse range of algorithms by restructuring their logical resources to reflect control or data characteristics of the algorithm. The hypercomputer architecture model is one such new proposal uniquely motivated not only to achieve high throughput but also to be easily programmable in a visually aided, object-oriented style. The underlying architecture is based on two salient features: an isotropic, MIMD array plane of thousands of simple, universal cells and a supporting 3D parallel-associative network which provides realtime IO through the planar dimension and control through the polar dimension. The programming model is integral to the architectural design and is based on an original formalism of algorithmically structured actor objects that maps to networks of hierarchically organized, cooperating cell groups. While hypercomputer architectural characteristics alone justify research into this area, only in the context of how we intend to program such a machine are the full implications of our ideas expounded. As such, the main contribution of this work is the application of reconfiguration and massive parallelism as a framework for marrying algorithms and architecture as well as programming and architecture.
- Research Organization:
- Polytechnic Univ., Brooklyn, NY (United States)
- OSTI ID:
- 5659713
- Country of Publication:
- United States
- Language:
- English
Similar Records
Implementation of a fully reconfigurable multimicroprocessor
Mapping of parallel algorithms to reconfigurable parallel architectures. Final report