基于深度学习算法的智能网卡数据流卸载模型  

A data stream offloading model based on deep learning algorithms

作　　者：赵武清 柏姗姗 李承钊 耿新 李科德 ZHAO Wuqing;BAI Shanshan;LI Chengzhao;GENG Xin;LI Kede(China Southern Power Grid Digital Grid Group Information and Telecommunication Technology Co.,Ltd.,Guangzhou 510663,China)

机构地区：[1]南方电网数字电网集团信息通信科技有限公司,广东广州510663

出　　处：《粘接》2024年第11期139-142,共4页Adhesion

摘　　要：为实现智能网卡数据流卸载优化的问题,在完全信息场景和不完全信息场景下,分别采用了不同的基于深度学习算法的智能网卡数据流卸载策略。结果表明,基于深度学习算法的完全信息场景下智能网卡数据流卸载策略中,运用Stackelberg博弈模型进行流量卸载策略研究,总流量负载、博弈方法总流量负载、对比方法总流量负载随着时段的变化规律基本相同,且在相同的时段下,流量从大至小顺序为总流量负载、对比方法总流量负载、博弈方法总流量负载。在不完全信息场景下,将Gradient Bandit算法用于在数据流分类和卸载中,总流量负载、强化学习方法总流量负载、对比方法总流量负载随着时段的变化规律基本相同。In order to realize the optimization of smart NIC data stream offloading,different data flow offloading strategies based on deep learning algorithm are adopted in the complete information scenario and incomplete information scenario.The results showed that in the complete information scenario of intelligent network card data flow offloading strategy based on deep learning algorithms,the Stackelberg game model was used to study the traffic offloading strategy.The total traffic load,the total traffic load of the game method,and the total traffic load of the comparison method changed with the same time period.Under the same time period,the order of traffic from highest to lowest was:total traffic load>comparison method total traffic load>game method total traffic load.In incomplete information scenarios,when the Gradient Bandit algorithm was used for data stream classification and offloading,the total traffic load,reinforcement learning method total traffic load,and comparison method total traffic load had basically the same variation pattern over time.

关 键 词：智能网卡 数据流 深度学习 信息场景 卸载模型 

分 类 号：TN929.5[电子电信—通信与信息系统] TP391.9[电子电信—信息与通信工程]