Long Short-Term Conditional Probability Matrix Prediction in Energy Efficient Ethernet

Abstract

Recently, much research on Energy Efficient Ethernet(EEE) have been carried out. For example, the recently proposed Frame Coalescing with Dynamic Threshold(FC-DT) strategy takes account of users’ target delay and can achieve good energy savings. This paper reveals the issue of large prediction error in FC-DT. To solve this problem, we propose FC-DT+ with improving the prediction algorithm. In addition, we enhance the FC-DT strategy by choosing the low power mode according to the actual data arrival rate when the target delay is within [2.75$\,\upmu$s, 4.44$\,\upmu$s]. Simulation results confirm that FC-DT+ has better prediction accuracy about 8% higher than that of FC-DT and accordingly has a better tail delay which is 1.5$\,\upmu$s lower than that of FC-DT on average. Moreover, FC-DT+ can reduce the energy consumption by 20–30% with the enhanced low power mode selection mechanism.

Y. Wang—Co-first author.