Dynamically Reconfigurable Automata for Estimating Interaction Between Data Flows in High-Performance Computing Systems

In High Performance Computing (HPC) systems, a large number of separate groups of tasks are executed in parallel. The computing and network resources of the HPC system should be provided to groups of tasks (distributed between tasks) in such a way that the required time characteristics are provided. There should be no interdependencies between the flows of transmitted data. The characteristics of data flows can change significantly over time. In order to ensure efficient utilization of network resources, their dynamic reconfiguration is required. It is necessary to ensure the absence of interdependencies between data flows, including in the process of dynamic reconfiguration. The paper proposes an approach based on dynamically reconfigurable automata to solve this problem.