Implementation Variants of the Global Distributed Associative Computing Environment for the Parallel Dataflow Computing System “Buran”
N. N. Levchenko, Anatoly Okunev, D.N. Zmejev · 2019
For the high-performance parallel dataflow computing system (PDCS) “Buran” this article proposes to move away from dividing the computational space into basic computational cores and moving on to organizing the system from associative computational cores. The use of the global distributed associative computing environment changes the architecture of the system and creates significant advantages compared with the basic PDCS architecture. The article discusses the implementation variants of the matching processor and the entire PDCS architecture in the form of a global distributed associative environment of information processing. The article describes the testing of the distributed associative environment in the behavioral cycle-accurate simulator, which was performed on various tasks. In the course of the experiments, the behavior of the new computing system architecture with hardware failures was also studied.