基于网络状态预测的提前控制  

PROACTIVE CONTROL BASED ON NETWORK STATE PREDICTION

作　　者：卢锡城[1] 朱培栋[1] 

机构地区：[1]国防科学技术大学计算机科学系,长沙410073

出　　处：《计算机研究与发展》1999年第9期1025-1031,共7页Journal of Computer Research and Development

基　　金：国家"八六三"高技术计划基金

摘　　要：高速网络的大带宽延迟乘积使基于传统反馈机制的 Ａ Ｂ Ｒ 难以对高可变 Ｖ Ｂ Ｒ 快速响应，造成速率的不匹配．为了提高反馈控制方法的有效性，充分利用高优先级 Ｖ Ｂ Ｒ 业务剩余的可用带宽，文中提出基于预测的提前控制模式，根据 Ａ Ｂ Ｒ 信源的不同反馈延迟对未来时间段 Ｖ Ｂ Ｒ 占用带宽进行预测，并利用未来 Ａ Ｂ Ｒ 可用带宽估算值为相应的 Ａ Ｂ Ｒ 进行速率分配，从而实现速率的精确匹配．高速网络 Ｖ Ｂ Ｒ 业务的强相关性使相对精确的时间序列预测成为可能，并可使用通用的线性预测方法完成 Ｖ Ｂ Ｒ 业务的在线实时预测；业务预测误差和 Ｒ Ｔ Ｔ 估计误差，可以根据当前负载因子（ Ｌ Ｆ）适当调整 Ｅ Ｒ 反馈值加以补偿．提前控制模式的有效性和公平性通过示例可以得到直观的认识。The large bandwidth\|delay product of high speed networks(HSNs) makes the conventional feedback\|based mechanisms unsuitable for ABR to fast respond to high bursty VBR traffic, and results in a significant rate mismatch . In order to increase the efficiency of feedback\|based control and acquire full access to left\|over bandwidth by high priority traffic, a prediction\|based proactive control scheme is proposed, which predicts future VBR bandwidth occupancy respectively for different feedback delays of ABRs, and assigns the corresponding ABR sources a rate on the basis of the estimate of total ABR capacity. Strong dependency of VBR traffic in HSNs makes possible the relatively exact prediction of time series, and we can take advantage of the existing general\|purpose linear prediction methods to get real\|time results on\|line. The current value of load factor(LF) can be used to adjust ER feedback , thus compensating the forecasting variance and RTT estimate error. The efficiency (i.e., exact rate match achieved) and fairness of the proactive scheme can be seen intuitively from an example, while their general description and simple proof are also presented.

关 键 词：提前控制 VBR业务 状态预测 计算机网络 ATM 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]