Low Power Design based on Neural Network Forecasting for Interconnection Networks

Power consumption is a key issue in high-performance interconnection network design. In interconnection network, power consumption of interconnection links will take up an ever larger portion of the power budget as demand for network bandwidth increases. Network traffic is an important factor that influences interconnection network power consumption. In this paper, we presented a traffic forecasting model using neural network method. Our interconnection network traffic forecasting model based on neural network adopted back-propagation learning algorithm and divided network period of time into two parts: stationary-hours and non-stationary-hours. Then we proposed voltage scaling algorithm for links based on our forecasting model. The results show that our forecasting model is well matched to real network traffic, and our links voltage scaling algorithm reduce power consumption of interconnection network effectively