基于门控循环神经网络的网络时延预测模型

A Network Latency Prediction Model Based on Gated Recurrent Neural Network

在信息中心网络ICN与5G融合的新型网络架构下,现场增强名字解析系统通过确定性时延名字解析服务来应对工业物联网IIoT等新应用对确定时延的挑战,其节点结构划分和维护需要测量节点间的时延。精确的网络时延预测可以减少测量代价,比简单移动平均方案更好地应对网络时延的变化。本文设计并实现了一种基于门控循环神经网络的网络时延预测模型,并基于亚马逊公司异地机房之间的真实时延数据进行了验证。实验结果表明,所提模型的预测精度比传统模型平均提高了20%以上,能够在业务场景中得到应用。

Under the new network architecture of ICN and 5G,the on-site enhanced name resolution system responds to the deterministic latency requirements from new applications such as Industrial Internet of Things(IIoT)through deterministic latency name resolution services.The division and maintenance of its node structure require latency measurements between nodes.Accurate network latency prediction can reduce the measurement cost and better respond to the changes in network latency than simple moving average solutions.This paper designs and implements a network latency prediction model based on gated recurrent neural network,and verifies it based on real latency dataset between Amazon’s remote computer rooms.The experimental results show that the prediction accuracy of the proposed model is improved by more than 20%on average compared with the traditional model,and it can be applied in business scenarios.