High Performance in Tree-based Parallel Architectures

Fabio Ancona, Stefano Rovetta and Rodolfo Zunino

D.I.B.E. - University of Genoa
Via all'Opera Pia 11a - 16145 Genova - Italy



ABSTRACT

The integrated scheme called "complex node" couples twin processors,
which share a dual-port memory supporting a bidirectional, high-speed
communication link. The paper considers the contribution of this device
to tree-structured processor hierarchies. As shown by theoretical analysis,
increased complex node performance may ensure optimality under reasonable
conditions in terms of connectivity and memory speed, which can be easily
attained by commercial equipment. This endows the overall research with
additional value from a technical point of view. The integrated approach
has been implemented by using transputers for experimental development;
related advantages in a specific application (implementation of associative
models) are also highlighted.


Proceedings of the 23rd EUROMICRO Conference
Budapest, Hungary, September 1997, pp. 474-481