Table 1.

Overview of the most relevant single-flow TCP CC algorithms.

CC Algorithm	Introduction Year	Type	Features
Tahoe [22]	1988	Loss-based	Slow start, congestion avoidance and a fast retransmit mechanism.
Reno [27]	1990	Loss-based	Introduced fast recovery mechanism.
NewReno [29]	1999	Loss-based	Fast recovery modification to allow multiple retransmissions.
SACK [30]	1996	Loss-based	Selective ACK option.
Westwood [31]	2001	Loss-based	Introduced a faster recovery mechanism that controls the sending rate according to the available bandwidth estimation.
HighSpeed [32]	2003	Loss-based	Introduced a modified TCP response function to allow for a faster cwnd increase and a faster recovery time in situations with a high cwnd size.
Scalable TCP [33]	2003	Loss-based	After each received ACK during the RTT, algorithm increases the cwnd size in proportion with the defined constants. After the packet loss decreases cwnd by a smaller factor, the standard CC is exploited.
BIC [34]	2004	Loss-based	Uses the binary search increase and additive increase techniques to determine the cwnd size.
CUBIC [35]	2008	Loss-based	Uses the cubic function for cwnd control characterised by a steady state and a maximal probing behaviour.
Vegas [36]	1994	Delay-based	Modification of TCP Reno that predicts network congestion before an actual loss of segments occurs. Uses a fine-grained RTT estimation and has a very efficient segment retransmission schedule.
Vegas+ [37]	2000	Delay-based	Modification of TCP Vegas that introduced an aggressive mode to overcome fairness issues when competing with TCP Reno.
New Vegas [38]	2005	Delay-based	Implemented three server-side modifications of TCP Vegas to overcome performance issues in a high latency environment.
Vegas-A [39]	2005	Delay-based	Implemented modified congestion avoidance mechanism to address fairness, rerouting and bias against high bandwidth connections issues of TCP Vegas in wired and satellite networks.
Vegas-V [40]	2012	Delay-based	Modification of TCP Vegas-A that addresses fairness and aggression issues when competing with TCP Vegas, TCP Vegas-A and TCP Reno flows.
FAST [41]	2004	Delay-based	Designed for high-speed long-latency networks. Adjusts the cwnd according to the feedback information of the average RTT and average queuing delay. Uses scaling parameters to effectively utilise the network capacity.
Compound [42]	2006	Hybrid	Based on the loss-based slow start phase. During the congestion avoidance phase, it uses a combination of two components, a standard loss-based and a new scalable delay-based component.
YeAH [43]	2007	Hybrid	Performs a dynamic exchange between a slow mode during the congestion avoidance phase and a fast mode during the fast recovery phase.
BBRv1 [44]	2016	Hybrid	Builds an explicit model of the network using the estimated RTT and the estimated available bottleneck bandwidth in order to prevent congestion.
BBRv2 [45]	2018	Hybrid	Uses ECN signals, improved fairness with CUBIC and lower packet loss rates for the optimisation of the TCP CC performance.