UDC 519.872
DEVELOPMENT AND INVESTIGATION OF MATHEMATICAL MODELS OF "PROCESSOR-MEMORY" SUBSYSTEM OF NUMA AND SUMA MULTIPROCESSOR SYSTEM ARCHITECTURES IN OPEN QUEUEING NETWORKS
A. I. Martyshkin, PhD (technical sciences), associate professor, department of computers and systems, PSTU, Penza; This email address is being protected from spambots. You need JavaScript enabled to view it..
In the article the problem of synthesis and investigation of models of «processor-memory» subsystem of multiprocessor systems is considered. The aim of this work is to study the architecture of «processormemory» subsystem of contemporary high performance computing, calculate and compare the characteristics obtained, as well as to get the evidence on the impact of conflict over access to shared resources on overall performance of the whole system. The objects of study in this work are «processormemory » subsystem of multiprocessor computing systems, existing types of this subsystem construction architecture. On the mathematical models offered the options for the implementation of «processor-memory» subsystem of multiprocessor computing systems with NUMA and SUMA are investigated. Models were
considered with the application of queuing theory. The analysis of influence of various factors on the mathematical model under consideration is made. The research was a comparative analysis of systems with NUMA and SUMA. Their strengths and weaknesses are identified. All necessary conclusions of this work are given. The considered models allow to evaluate the performance of multiprocessor systems without building a real layout. Due to this the economic effect is achieved since the evaluation of designed system characteristics and the choice of optimal variants can be performed without building a real system.
Key words: multiprocessor computing system, system architecture, mathematical model, performance, queuing system, NUMA architecture, SUMA architecture, probabilistic-temporal characteristics of computing systems.