An Efficient Framework of Network Performance Estimation

Network performance estimation is a key enabler in achieving efficient network operation. In this paper, we propose a novel framework Flownet for generalized network performance estimation with a graph neural network and a pioneering transformer to learn and model graph-structured information at the flow level. Our extensive experiments have shown that Flownet can improve the MAPE by up to 65.18% compared with the state-of-the-art models in diverse scenarios.