A Traffic Model For Networks-on-chip In Heterogeneous SoC

In the study of the Networks-On-Chip(NoC),NoC's traffic characteristics have a great influence on the network performance.It is of great significance to research the NoC traffic model.The existing NoC traffic model can be divided into two kind,one is a short correlation model and the other is for homogeneous multi-core System-On-a-Chip(SoC)research.Traffic model for heterogeneous multi-core system is very few.Compared with the homogeneous SoC,processor cores in the heterogeneous SoC is not only the Central Processing Unit(CPU)as well as Graphics Processing Unit(GPU).It is necessary to model the traffic on different processor modules according to the application.In this thesis,a traffic model for heterogeneous multi-core SoC is proposed.The original traffic is extracted from the NoC interfaces using the ARM DS-5 streamline and then the traffic model with three parameters is composed according to the prominent characteristics of the traffic.The three parameters are mean value,variance and self-similarity.In the different application scenarios of heterogeneous SoC,the values of the three parameters are different.In the different application scenarios,the traffic in different IP modules is different obviously.GPU injects more traffic in the application scenarios needed by GPU such as graphics and image processing applications and the traffic shows high bandwidth characteristics.The characteristic parameters in the traffic model can be fitted by fractal Gaussian noise to generate self-similar sequence.The original traffic has the same self-similarity as the self-similar sequence and can be instead by the self-similar sequence.Experiment results show that the error rate is less than 5%compared with the Hurst parameter of the original traffic and the self-similar sequence.Using the fitted self-similar sequence and the original traffic as the NoC emulator traffic input respectively,the error rate of the NoC performance is also less than 5%with the same simulation time.