基于最小均方滤波的RTO预测算法  

RTO prediction algorithm based on a least mean square filter

作　　者：陈翔[1] 刘卫东[1] 任丰原[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《清华大学学报（自然科学版）》2007年第4期603-605,共3页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60573122);国家"九七三"重点基础研究项目(2003CB314804)

摘　　要：传统的传输控制协议(TCP)中的重传超时值(RTO)设置过于保守,且RTO预测算法对于往返时延(RTT)变化的响应过于缓慢,造成长时间的无谓等待,降低了带宽利用率。为了解决这些问题,提出一种基于最小均方(LMS)滤波的预测算法。为了验证新算法的性能,选择了大量有代表性的网址进行试验,对新旧2种算法的预测结果进行比较,并做了性能分析。试验结果证明,所提出的算法相比于原算法,具有能很快达到稳定,稳定后的预测误差小,对RTT变化的响应快等特点,提高了传输性能。The retransmission timeout （RTO） prediction algorithm that current the transmission control protocol uses is too sluggish to respond to changes in the round-trip time （RTT）. Furthermore, the predicted values are usually much larger than real RTTs, leading to reductions in the bandwidth utility ratio. A prediction algorithm based on a least mean squares filter was developed to predict better values. The algorithm performance was evaluated using a variety of typical websites. The results show that the algorithm quickly stabilizes, has a very small prediction error, and quickly responds to RTT changes, so it can improve transmission performance.

关 键 词：往返时延(RTT) 重传超时值(RTO) 最小均方误差(LMS) 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]