Congestion Control Based On Explicit Feedback

Congestion control is a resource allocation and scheduling problem,whose core goal is to allocate network bandwidth to the network end host reasonably,and the rationality is embodied in two aspects.One is that the bandwidth resource is fully utilized and does not cause network congestion,and the other is the relatively fair bandwidth share of different end hosts.Around the core goal of congestion control,a series of congestion control algorithms were designed.In principle,it is divided into three categories.The first one is judging whether there is a congestion based on packet loss,the second one is based on delay variation,and the third conduct congestion control based on explicit feedback of the router.This paper mainly focus on the third one.This paper first analyzes the representative algorithms of traditional congestion control algorithms,such as TCP-Cubic,TCP-Vegas and TCP-BBR.Through experimental data,it is found that both the packet loss based congestion control algorithm and the delay based congestion control algorithm have some limitations,including the sensitivity to random packets and easy to cause additional delay,poor fairness,and so on.This paper analyzes the representative algorithm of explicit congestion control,including XCP,RCP,DCTCP,VCP and so on.The experimental data shows that this kind of algorithm can make full use of high speed network and do not cause additional delay,but the current algorithms are too complex to be deployed and easy to cause burst traffic.Based on the analysis of current research of explicit congestion control,the BCTCP protocol is designed for the purpose of improving link utilization and reducing delay.It is based on the load factor of router.Three phase window adjusting strategy of MI-AI-MD is adopted,Current Window Validity and Pacing are introduced,which improves the link utilization in different network scenarios such as high packet loss scenarios and very bursty scenarios,and keep the size of router's queue in control.In this paper,to improve the fairness convergence speed of current algorithms,we analyzes the principle of fairness convergence of MIMD strategy,and designs a router end fairness controller based on Bloom filter.After the router adopts the controller,different load factor will be fed to the end host according to their throughput.Based on this load factor,fairness convergence speed of congestion control algorithms can be improved.Based on this algorithm,the SFECN protocol is designed.Finally,for the convenience of future deployment,this paper attempts to use the CE bit of the standard ECN header to carry out the explicit congestion control.We proposed One Bit SFECN,which reaches the effect close to the SFECN,because it no longer needs to modify the existing message format of TCP/IPV4 protocol,it lower the cost of future deployment.