The implementation of the computational graph for parallel computing

Chaikovsky D. S., Gulevich N. A., Pchelinceva E. G.

The subject of research is the analysis of software implementations of a computational graph in C++ using the Cpp libraries-Taskflow, Task dependency of the OpenMP standard and Intel Threading Building Blocks (TBB). The features of the software implementation of the computational graph using the open source library Cpp-Taskflow, which is written for C++, are considered. The relevance of the research is determined by the wide spread of big data processing technologies, parallel programming and the need to study and create appropriate tools, including for the software implementation of computational graphs. The paper identifies the main limitations faced by developers of parallel programs. Conclusions are made that the implementation of a computational graph in Cpp-Taskflow has advantages over existing tools.

Keywords:parallel computing, parallel programming, computational graph, dataflow architecture, Cpp-Taskflow library, parallel programming on Cpp, object-oriented programming, OpenMP, Intel TBB, methods describing parallelism.