Towards Hybrid Computing with Scalable Mass Parallelism for High-Performance Cloud Computing of Digital Twins

In order to improve the computing performance of the cloud implementations of digital twin models, the possibilities of hybridization of cloud servers using single-chip accelerating general-purpose computers with mass multi-core parallelism are investigated. Such accelerating computers belong to the class of Multi-Processor Systems-on-a-Chip (MPSoC). The variants for system-engineering integration of mass multi-core parallelism of accelerating computer of class MPSoC into a virtual environment of cloud computing resources are considered. On the example of the requirements for the creation of digital twins of nuclear power plants (NPPs), the possibilities of using hybrid computing systems based on the proposed single-chip accelerator with mass multi-core parallelism of the MPSoC class are considered. The proposed multi-SIMD architecture of such accelerators is based on the scalable multiprocessor architecture of accelerating computers PS-2000 for the eponymous hybrid computing complexes, which were produced in a large industrial series in the absence of world analogues. The high efficiency of architecture PS-2000 had been confirmed by the wide practice of application in various fields. At the conceptual level, it is shown that the use of the proposed single-chip accelerating computers with in-chip integration of mass multicore parallelism into the virtual environment of cloud servers opens up opportunities for cloud systems to expand the scaling ranges of computing performance of cloud servers by orders of magnitude. It is necessary when building digital twins of technogenic systems with increased security and computing performance requirements, for example, for NPP.