Global Distributed Associative Environment - Evolution of Parallel Dataflow Computing System “Buran”

Abstract

A high-performance parallel dataflow computing system that implements a dataflow computing model solves problems in the field of supercomputing, the main one of which is low real performance when parallelizing tasks for hundreds of thousands of computational cores. In this article, a new architecture for the parallel dataflow computing system is proposed - a global distributed associative environment. The concept of the computing system organization is described, which consists in dividing the system into associative computational cores, which in turn consist of a set of local associative elements. The new architecture demonstrates significant advantages over the previous architecture, primarily due to the increased reliability of the system and the simplification of programming.