基于流量预测的日志写入方法  

Log writing method based on traffic prediction

作　　者：齐实 黄小红 马严 丛群 QI Shi;HUANG Xiaohong;MA Yan;CONG Qun(Information Network Center,Institute of Network Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;Beijing Wrdtech Co.,Ltd,Beijing 100876,China)

机构地区：[1]北京邮电大学网络技术研究院信息网络中心,北京100876 [2]北京网瑞达科技有限公司,北京100876

出　　处：《通信学报》2018年第A01期129-134,共6页Journal on Communications

基　　金：教育部-中国移动科研基金资助项目(No.MCM20160304)~~

摘　　要：为了解决校园网络日志存储中流量变化导致的机械硬盘平均访问时间长、波动大的问题,设计了一种通过流量预测进行日志写入的方法。首先,提出流量定界算法对流量进行划分为高峰期、中间期与低谷期。然后,利用BP神经网络对不同时期的流量数据进行训练,构建流量预测模型。最后,基于流量预测模型,设计日志存储算法。实验结果表明:与已有的日志写入方法相比较,基于流量预测的日志写入方法在流量高峰期时机械硬盘的平均访问时间从90 ms减小至59 ms以下,同时,平均访问时间的方差从224.5降至50.2。In order to solve the problem of long average access time and large fluctuation of hard disk caused by trafficchanges in campus network log storage,a methodwith log writing through traffic prediction was designed.First,a traffic partition algorithm to divide traffic to three periods was proposed:peak period,intermediate period and low period.Then,BP neural network was used to train the traffic data to construct a traffic prediction model.Finally,based on the traffic prediction model,a log writingalgorithm was designed.The experimental results show that compared with the existing log writing method,the average access time of hard disk is reduced from 90 ms to less than 59 ms during the peak period,and the variance of the average access time is reduced from 224.5 to 50.2.

关 键 词：流量预测 日志存储 写入方法 平均访问时间 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]