Self-similar traffic study of on-chip interconnection networks

The objective of this paper is to study the self-similarity of on-chip interconnection network traffic. Traffic modeling with self-similarity can help designers obtain the performance characteristics of the on-chip network. This paper studies the relationship between the Traffic Distribution Graph (TDG) and the self-similar traffic parameter, Hurst parameter. We also use experimental data to find the fitting functions to estimate the Hurst parameter with TDG attributes. We adopt an example to evaluate the feasibility of the fitting functions. Simulation results show that our method can provide an effective traffic model for typical DSP systems in light of the deviation effect of the estimation method.

I. Introduction

With the rapid advance of VLSI technology, hundreds and thousands of processing cores can be integrated into a single chip. Nevertheless, the conventional bus based communication systems can handle only one pair of data communication simultaneously, and it has become a bottleneck for on-chip interconnections. Therefore, Network on Chip (NoC) is proposed as the promising multi-core interconnection technology [1].