Fragmentation of supercomputer resources is usually considered a negative factor that reduces the utilization of HPC systems. Most research focuses on minimizing the downtime of supercomputer nodes and cores by reducing fragmentation. This article presents a method for increasing computing speed based on the artificial fragmentation. Artificial fragmentation adds supercomputer nodes to certain jobs in order to reduce competition for resources among parallel program threads. The results of simulation modeling and experiments show that artificial fragmentation not only increases the speed of individual jobs, but also improves the performance of the supercomputer system when processing a jobs flow.
Baranov et al. (Sat,) studied this question.