Adaptive explicit congestion control based on bandwidth estimation for high bandwidth-delay product networks

Adaptive explicit congestion control based on bandwidth estimation for high bandwidth-delay product networks
复制标题

DOI:
10.1016/j.comcom.2013.02.006
复制
发表时间:
2013-06
期刊:
Comput. Commun.
影响因子:
--
通讯作者:
Jianxin Wang;Pingping Dong;Jing Chen;Jiawei Huang;Shigeng Zhang;Weiping Wang
Jianxin Wang;Pingping Dong;Jing Chen;Jiawei Huang;Shigeng Zhang;Weiping Wang
中科院分区:
其他
文献类型:
--
作者:
Jianxin Wang;Pingping Dong;Jing Chen;Jiawei Huang;Shigeng Zhang;Weiping Wang

文献摘要

被引文献

相似文献

现有的高带宽延迟积(HBDP)网络拥塞控制协议要么因网络状态反馈不足而导致效率下降,要么难以在实际网络中部署,因为它们需要修改IP报头以提供足够的反馈信息。基于VCP,本文提出了VCP-BE协议,该协议可以通过自适应带宽估计器估计发送方的可用带宽来提供更准确的网络状态信息。估计器可以动态调整其参数(例如观察间隔)以实现不同的精度和响应能力,以适应由两个ECN位指示的网络状态,而无需比VCP更多地修改IP标头。此外,当网络负载较高时,VCP-BE计算当前吞吐量与可用带宽的比率来判断网络是否接近过载状态。通过这些细粒度的网络状态估计,VCP-BE可以比VCP更精确地调整拥塞窗口,从而大大提高其收敛速度,实现高带宽利用率和公平性。仿真结果表明,VCP-BE 的性能也优于 MLCP,MLCP 使用 7 位进行显式反馈,而 VCP-BE 仅使用 2 个 ECN 位。
Existing congestion control protocols for high bandwidth-delay product (HBDP) networks either suffer from efficiency degradation caused by insufficient feedback of the network state, or are difficult to deploy in real networks because they require modification in IP header to provide enough feedback information. Based on VCP, this paper proposes the VCP-BE protocol that can provide more accurate network state information by estimating available bandwidth for sender with an adaptive bandwidth estimator. The estimator can dynamically adjust its parameters (e.g., observing interval) to achieve different accuracy and responsiveness to adapt to the network state indicated by two ECN bits, without no more modification in IP header than VCP. Furthermore, when the network load is high, VCP-BE calculates the ratio of current throughput to the available bandwidth to judge if network is close to the overload state. With these fine-grained network state estimation, VCP-BE can adjust the congestion window more precisely than VCP, thus greatly improves its convergence speed of achieving high bandwidth utilization and fairness. Simulation results show that VCP-BE also outperforms MLCP, which uses seven bits for explicit feedback while VCP-BE only uses two ECN bits.