Abstract: With the continuous development of computer network and the enhancement of network performance requirements, it is difficult to meet the people’s needs for traditional network. OpenFlow makes up for traditional network. However network session identification is inefficient and packet forwarding path selection is poor. Focusing on forwarding path and matching, this paper proposes GODP (GPU OpenFlow data processing) model. The GODP model combines GPU with biological sequence algorithms and machine learning methods, and presents the GPTWF network session matching algorithm and network session forwarding algorithm to accelerate matching speed and improve network environment. The experiments show that network session matching algorithm gives a speedup 290, and network session forwarding algorithm makes link loss rate less than 5%, with an average decline 62.71%, and network delay less than 20 ms, average decline 73.88%.  

Key words: OpenFlow technique, GODP model, GPU computation, biological sequence, machine learning

摘要： 随着计算机网络的不断发展以及人们对网络性能要求的不断提高，现有网络很难满足人们的需要。OpenFlow的出现能够很好地解决现有网络的不足，但存在网络会话识别效率不高，网络报文转发路径选择不佳等问题。针对匹配算法和路径转发进行了研究，提出了GODP（GPU OpenFlow data processing）模型。该模型通过融合GPU计算与生物序列算法和机器学习方法，提出了GPTWF网络会话匹配算法和网络会话转发算法，有效提升了匹配效率，优化了网络环境。实验表明网络会话匹配算法加速比提升近290，网络会话转发算法使得链路丢包率低于5%，延时小于20 ms，网络会话丢包率和延时分别平均下降62.71%和73.88%。

关键词: OpenFlow技术, GODP模型, GPU计算, 生物序列, 机器学习