A note on the elementary execution unit in a parallel inference machine
When we design a parallel inference machine which executes logic programs, there are some important problems that greatly influence the machine performance. The most important problem is how to decide the elementary execution unit (granule). The less the size of the elementary execution unit is, the less is the overhead of copying and transferring, but the control among the units becomes more complicated.
Until now, a few models of parallel inference machines have been proposed. However, we think that this problem is not discussed enough in these models. In this paper, we discuss the problem in detail.
KeywordsInternal Representation Logic Program Machine Performance Sequential Machine Structure Skeleton
Unable to display preview. Download preview PDF.
- Haridi, S. and Ciepielewski, A., An OR-parallel Token Machine, Logic Programming Workshop 83, 1983.Google Scholar
- Ito, N. and Masuda, K., Parallel Inference Machine Based on the Data Flow Model, Proc. of International Workshop on Highlevel Computer Architecture 84, 1984.Google Scholar
- Moto-oka, T., Tanaka, T. et al, The Architecture of a Parallel Inference Engine-PIE-, FGCS '84, ICOT, 1984.Google Scholar
- Onai, R. et al, The Architecture and Software Simulation of the Parallel Inference Machine PIM-R, ICOT Technical Report TR-077, 1985 (in Japanese).Google Scholar
- Yuhara, M. et al, A Unify Processor Pilot Machine for PIE, Proc. of the Logic Programming Conference '84, Tokyo, 1984.Google Scholar