Abstract
Abstract
Finding the optimal strategy in network routing has always been a NP hard problem. Due to the complexity and dynamics of network traffic, the existing intelligent routing algorithms have poor generalization ability. Therefore, this paper proposes an intelligent routing strategy based on deep reinforcement learning, and with the help of SDN control can dynamically collect network traffic distribution information, can dynamically adjust the routing strategy. Compared with traffic engineering algorithms such as TCMP and DRL-TE, the end-to-end delay is optimized under different throughput.
Subject
General Physics and Astronomy
Reference17 articles.
1. The cooperative associate for Internet dataanalysis (CAIDA)
2. A survey onquality-of-service routing algorithms for the Internet[J];Cui;Journal of Software,2002
3. Deep Architecture for Traffic Flow Prediction: Deep Belief Networks with Multitask Learning[J];Huang;IEEE Transactions on Intelligent Transportation Systems,2014
4. The deep learning vision for hetero-geneous network traffic control:proposal, challenges, and future perspective[J];Kato;IEEE Wireless Communications,2017
5. Network traffic classification using machine learning techniques over software defined networks[J];Parsaei;International Journal of Advanced Computer Science and Applications,2017