Table 2.

Review of the research articles on CC algorithms in 5G networks.

Simulation Scenarios	Evaluated TCP CC Algorithms	Main Algorithm Drawbacks	Summary
UE experiencing LOS to NLOS transitions and outage events between mmWave BS in small buildings and large building scenarios [76].	NewReno	Large buffer: bufferbloat. Small buffer: early buffer overflow. Long outage: throughput degradation and slow throughput recovery.	Evaluated loss-based CC algorithms showed slow reaching of full throughput, large data rate drops, increased latency and slow throughput recovery.
	CUBIC	Long outage: throughput degradation and slow throughput recovery.
UE experiencing blockages from other humans and from buildings [78].	Cubic with AQM CoDel	Human blockage: packet drops, slow throughput recovery. Building blockage: multiple packet drops resulting in near-zero throughput.	AQM CoDel does not mitigate the bufferbloat problem and DRW showed a much higher throughput and negligible oscillation in the delays.
	CUBIC with DRW	Low delay and much higher throughput in both scenarios.
High-speed train scenario with different buffer sizes and a dense urban scenario, using remote server and edge server deployment [13].	NewReno	Remote server: lowest goodput.	Latency is greatly reduced for all observed CC algorithms using edge server deployment. Applying the AQM scheme with loss-based CC algorithms can reduce the latency in large buffer deployments.
	CUBIC	Edge server: lowest goodput.
	HighSpeed	Big buffer: high latency and goodput.
	BBR	Big buffer: high latency and goodput. Small buffer: low latency with weak goodput reduction.
UE experiencing blockages between mmWave BS in extensive blockages, medium blockages and multiple short blockages scenarios. Handover scenario between three BSs and a mobile user experiencing multiple short to extensive blockages. Dense small cell deployment with various obstacles in a situation of multiple BSs serving multiple UEs when short flows and background traffic coexist [12].	NewReno	Blockage events: Slow full throughput reach after multiple losses and slow network probing in the congestion avoidance phase.	Blockage events greatly impact latency for loss-based CUBIC and Scalable TCP. Delay-based Vegas showed the lowest throughput with minimal latency variability. Hybrid CC algorithms showed minimal performance variations. Loss-based CUBIC showed high-performance variations in longer NLOS periods as opposed to hybrid YeAH which showed minimal throughput variations and required fewer transmissions, but achieved less throughput compared to CUBIC.
	CUBIC	Blockage events: High RTT variability in LOS-NLOS transitions. Handover: fast throughput recovery from the slow start. Multiple flows: high number of retransmissions and high buffer occupancy, high throughput.
	Scalable TCP	Blockage events: High RTT variability in LOS-NLOS transitions.
	Vegas	Blockage events: Low throughput with minimal RTT variability.
	Westwood	Blockage events: Slow network probing in congestion avoidance phase.
	YeAH	Blockage events: Low RTT and minimal performance variability. Handover: slow throughput recovery from slow mode. Multiple flows: low number of retransmissions and high robustness.
	BBR	Blockage events: Low RTT in all scenarios and minimal performance variability.
Multiple BSs serving multiple vehicles moving at random speed in the mmWave CVNs environment using two different mobility models in rural and urban areas [79].	CUBIC	High cwnd size variability.	Due to the high channel fluctuations caused by mobility in CVNs, the RTW-TCP outperformed the existing CC algorithms as they cannot distinguish between congestion and link failures.
	Compound	High average RTT, lowest aggregate throughput and high cwnd size variability.
	X-TCP	Low average RTT and high cwnd size variability.
	RTW-TCP	Low throughput reduction due to mobility, low RTT and cwnd continued to increase despite blockages.
Multiple UEs communicating with single mmWave access point under static link, short blockages, long blockages, and mobility and blockages scenarios [80].	CUBIC	Long queuing delay and good fairness.	CUBIC showed a dramatic increase in the delays in NLOS conditions. BBR is not suitable for uninterrupted high-speed applications and Prague has fairness issues.
	BBR	Low queuing delays and good fairness. Periodically reducing sending rate.
	Prague with DualQ, AQM and AccECN	Lowest queuing delay and poor fairness.
Single gNB serving mobile users in a small building and large buildings scenario [81].	NewReno	Lowest performance.	D-TCP using cross-layer implementation to obtain SINR information showed the best performance among the evaluated CC algorithms.
	BIC	Relatively fast achieves full throughput.
	CUBIC	Long network probing.
	BBR	Relatively fast achieves full throughput.
	D-TCP	The best performance and almost instantly achieves full throughput.