A Flexible Analytic Model for the Design Space Exploration of Many-Core Network-on-Chips Based on Queueing Theory
Erik Fischer, Albrecht J. Fehske, Gerhard Paul Fettweis · 2012
Abstract—A continuing technology scaling and the increasing requirements of modern embedded applications are most likely forcing a current multi-processor system-on-chip design to scale to a many-core system-on-chip with thousands of cores on a single chip. Network-on-chip emerged as flexible and high-performance solution for the interconnection problem. There will be an urgent need for fast, flexible and accurate simulation models to guide the design process of many-core system-on-chip. In this paper, we introduce a novel analytic approach for modeling on-chip networks to fulfill these requirements. The model is based on queueing theory and very flexible in terms of supported topology, routing scheme and traffic pattern. The approach overcomes the limitations of the mean value analysis introduced in the existing work. Instead, it provides information about a steady-state distribution of the network routers. This allows to dimension network resources, such as buffers, links, etc. We show the high accuracy of the model by comparison with a cycle-accurate simulation. The model is able to estimate the mean network latency with an accuracy of about 3%. Keywords-network-on-chip; noc; queueing theory; analytic model. I.