Study of Execution Efficiency of Implementation Versions of Sparse Matrices Multiplication Algorithm on Parallel Dataflow Computing System “Buran”

Abstract

The article proposes to solve problems related to the parallel implementation of the sparse matrices multiplication task, using the architecture of the parallel dataflow computing system “Buran”, which implements the dataflow computing model. The article describes the implementation versions of the sparse matrices multiplication task algorithm in the dataflow programming paradigm. These algorithm implementations demonstrate the simplicity of their creation and universality. The experiments conducted using the behavioural cycle-accurate simulator have shown that the increase in the efficiency of tasks that use a sparse data structure can reach several orders of magnitude when executing them on the parallel data flow computing system.