Mitigating Bufferbloat With Receiver-Based Tcp Flow Control Mechanism In Cellular Networks
Xiaolan Liu,Fengyuan Ren,Ran Shu,Tong Zhang,Tao Dai
DOI: https://doi.org/10.1201/9781003337096-5
2017-01-01
Abstract:The symbol of network congestion is the packet dropping event detected by TCP congestion control mechanism. But in cellular networks, the packet dropping event is always concealed by a large buffer deployed necessarily to absorb the traffic burst and to achieve the retransmission mechanism of link layer for every user in the base station. Excessive packets beyond link capacity are blocked in the large buffer, so sender cannot detect the network congestion and will increase its sending window consistently. And this may trigger excessively long delays and network jitter without contribution to the improvement of throughput at the same time. This intractable phenomenon is named bufferbloat. More seriously, the dynamic variation of wireless link bandwidth can deteriorate such circumstance and severely degrade the network performance and user's QoE. Wireless link is always the bottleneck link of a TCP connection in cellular networks, and its available bandwidth is that of the connection. Wireless link is also the last hop of the connection, and it connects the mobile terminal directly. Thus, the mobile terminal can get all of the channel state information such as signal strength. So it is more feasible to resolve the bufferbloat at receiver side with the help of channel state information; nevertheless, which is ignored in previous studies. So in this chapter, we unveil the root cause of bufferbloat is the in-adaptation between the adjustment of sending windows and the dynamic variation of wireless link bandwidth at the existence of large buffer. And we present a receiver-based flow control strategy named ABRWDA that retrieves the available bandwidth of a wireless link at receiver side directly and uses it to dynamically calculate the receiving window (rwnd) and then adjusts sending windows to mitigate the bufferbloat in cellular networks. Because of the dynamics of wireless link bandwidth, we introduce kalman filter algorithm to maintain the system stability. We test our approach with the NS-2 simulation, and the results indicate that ABRWDA achieves 0.1 times and 0.4 times shorter queues, 0.5 times and 0.8 times lower latency, while still maintaining the same high throughput as that of Newreno and a previous solution DRWA, respectively.