Next: Modeling of Autonomous
Up: Projects
Previous: How the Brain Works
- Nikolay N. Mirenkov,
- Professor, Distributed Parallel Processing Laboratory
Main results obtained in 1995:
-
The set of optimal parallel array processors for fast solution of system of
linear equations was investigated. We first developed the efficient
highly-parallel versions of division-free Gaussian elimination method and
then the set of admissible array processors were synthesized and analyzed.
A new parallel algorithm for Householder bidiagonalization on parallel
computers with dynamic ring architecture was investigated. The experimental
study on CM-5 supercomputer shows that this parallel algorithm achieves high
speedup for large matrices.
-
We developed efficient routing algorithms for
hypercubes and star graphs with path-shaped obstacles.
We also presented a new concept for fault-tolerance routing,
namely, $d$-separated paths. Two paths are $d$-separated if their
distance is greater than or equal to $d$.
-
Approaches to the automatic synthesis of parallel programs on the
basis of visual specification for problems on grids of pixels were
developed. Though there are literally thousands of algorithms there
are very few design techniques. We distinguish these techniques
according to multi-dimensional structures used for algorithms, partial
orders of scanning nodes in structures and possible parallel
implementations. In this case visual specification subsystem supports a set
of parameterized computational schemes on 2-D structures where partial orders
of node scanning are defined by external configurations of surrounding nodes or
internal configurations (variable values) of nodes. It provides a
good basis for very-high-level parallel programming and parallel
program transparency. The synthesis of programs consists of two steps
and one of them is performed during run-time. This synthesis is based
on a set of the hand-made parallel procedures and some other knowledge of
application methods as well as on the analysis of initial data.