机构地区: 中山大学
出 处: 《计算机学报》 2004年第1期42-51,共10页
摘 要: 该文针对自相似的网络条件提出了一种单播的TCP Friendly拥塞控制算法 ,称为RAAR MT .该算法通过在线预测未来流量水平的变化趋势 ,并利用预测结果在接收端对业务流进行多时间尺度的速率控制 .仿真结果表明 ,在自相似的网络条件下该算法能够有效降低业务流的丢失率 ,即使在传统短相关环境中 ,其性能也不会出现严重衰减 .通过与TFRC协议进行对比研究发现 ,RAAR MT协议在对TCP的友好性、协议内的公平性以及速率的平滑性等主要的TCP Friendly协议指标都具有更好的性能 .由于RAAR MT不需要进行每包反馈 ,且主要功能在接收方实现 ,因此可方便地将该机制引入多媒体组播传输系统中 . Since the traffic in modern telecommunication networks always shows self-similarity or long-range dependence, a new unicast TCP-Friendly protocol using in self-similar traffic network, called RAAR-MT, is proposed in this paper. RAAR-MT on-line predicts the tendency of traffic level in the near future and use this predicted information to exert multiple time scale traffic control at receivers. Simulation shows that, comparing with RAAR protocol, RAAR-MT protocol can enhance the tracking ability to the network state and gain 10%-20% performance improvement in loss rate on self-similar traffic network. Even in traditional short-range dependent environment, the performance of this protocol will not degrade severely. Comparisons have been made between RAAR-MT and TFRC too. Results are that RAAR-MT has better performance in TCP-Friendliness, intra-protocol fairness and smoothness in typical self-similar network environment. RAAR-MT is also a promising avenue of development for congestion control of multicast multimedia traffic, since it is not a per-packet feedback and is mainly implemented at receivers.