QUERIES SERVICE TIME RESEARCH AND ESTIMATION DURING INFORMATION EXCHANGE IN MULTIPROCESSOR SYSTEMS WITH “UNI BUS” INTERFACE AND SHARED MEMORY
Keywords:
Buffer Element, Memory Latency, Memory Controller, Memory Bandwidth, Memory Architecture, “ProcessorMemory” Subsystem, Multiprocessor System, Record-Operation, Read-Operation, Transaction, Queueing Network System, Imitation Model, Analytical Model, SimulationAbstract
: The issues connected with estimating service time of queries (transactions) during the information exchange in multiprocessor systems with a unibus interface and shared memory are analyzed and studied in the article. The article aims at developing and making research of models based on systems and queueing networks, the "processor-memory" subsystem, as well as estimating the queries service time during the information exchange in multiprocessor systems with shared memory. The subject matter of the study is the analysis of time delays associated with conflict situations occured during the realization of interprocessor exchange when many processors turn to the exchange unibus and memory. The object of the article research is the "processor-memory" subsystem of existing multiprocessor systems and well-known versions of the architecture of this subsystem . The main task defined by the authors of the scientific article is to develop and make research of mathematical models of the "processor-memory" subsystem of the mentioned systems and to estimate the processing time of inputting queries during the information exchange in systems with shared memory. Mathematical models for carrying out queries service time research have been proposed. Equations for determining the main probabilistic-temporal characteristics of the "processormemory" subsystem have been presented. The mentioned probabilistic-temporal models have been developed using the theory of queueing networks and probability theory. In conclusion the authors make the main judgements about the work done. The mathematical models studied in the article make it possible to estimate the main probabilistic-temporal characteristics of multiprocessor systems without developing real models or prototypes. As a result some effect is achieved, because it is possible to estimate the characteristics of new multiprocessor computer systems and choose the most optimal ones without creating a real expensive system.
