Parallel Simulation Methods of Heteroepitaxial Growth on Multiprocessor Systems with Shared Memory

This paper proposes solutions that increase the efficiency of parallel program execution on high-performance computing systems when simulating heteroepitaxial growth on multiprocessor systems with shared memory. The developed algorithms aim to implement in software the simulation of heteroepitaxial growth on multiprocessor NUMA nodes with shared memory. The primary prerequisite for the effective execution of parallel programs on the resources of a multiprocessor node is the incorporation of an architectural approach to the implementation of algorithms for data transfer through the shared memory of NUMA nodes. The proposed algorithms for optimizing synchronization in shared memory systems enhance the efficiency of accessing the shared memory of a multiprocessor node and reduce the time required for barrier synchronization. The developed methods and algorithms are implemented in the form of software for multiprocessor NUMA nodes with shared memory.