Simulation Model for Controlling the Processing of Information Packets by a NoC Router

This article describes the challenges of modeling switch fabric traffic that is part of NoC routers. Simulation is often used to validate a NoC design and identify potential performance bottlenecks before implementing hardware. Due to the distributed and complex network topology of routers, arbitration and routing algorithms, etc., timing, buffer queuing, and hence infrastructure performance are difficult to predict. Network traffic also generates application traffic in data processing nodes. When processing NoC traffic, the activity of IP cores is unstable and does not have stationarity, so it is very important to find traffic control mechanisms. Often, well-known models built on the basis of queuing theory do not always correctly reflect the connection request processing logic, do not take into account deadlocks and other features of the system functioning. An approach that can more accurately take into account the nuances of the switch is to build an agent-based model. To create it, we used the Mesa framework implemented in the Python programming language. The switch has small built-in buffers - one for each input. The model used a switch with 5 inputs and 5 outputs. The idea behind switch traffic management is to count how often there are transacts in the internal buffers that are directed to one exit. If they occur frequently enough, then the information flow is split and two or more outputs are given the same identifier to service such transacts in parallel.