Evaluation Study for Delay and Link Utilization with the New-Additive Increase Multiplicative Decrease Congestion Avoidance and Control Algorithm

As the Internet becomes increasingly heterogeneous, the issue of congestion avoidance and control becomes ever more important. And the queue length, end-to-end delays and link utilization is some of the important things in term of congestion avoidance and control mechanisms. In this work we continue to study the performances of the New-AIMD (Additive Increase Multiplicative Decrease) mechanism as one of the core protocols for TCP congestion avoidance and control algorithm, we want to evaluate the effect of using the AIMD algorithm after developing it to find a new approach, as we called it the New-AIMD algorithm to measure the Queue length, delay and bottleneck link utilization, and use the NCTUns simulator to get the results after make the modification for the mechanism. And we will use the Droptail mechanism as the active queue management mechanism (AQM) in the bottleneck router. After implementation of our new approach with different number of flows, we expect the delay will less when we measure the delay dependent on the throughput for all the system, and also we expect to get end-to-end delay less. And we will measure the second type of delay a (queuing delay), as we shown in the figure 1 bellow. Also we will measure the bottleneck link utilization, and we expect to get high utilization for bottleneck link with using this mechanism, and avoid the collisions in the link.

💡 Research Summary

The paper addresses the growing importance of congestion avoidance and control in an increasingly heterogeneous Internet. It focuses on three key performance metrics—queue length, end‑to‑end delay, and bottleneck link utilization—when evaluating congestion control mechanisms. Building on the classic TCP Additive Increase Multiplicative Decrease (AIMD) algorithm, the authors propose a modified version called New‑AIMD. The central idea of New‑AIMD is to make both the increase and decrease phases adaptive to current network conditions. Instead of using a fixed multiplicative decrease factor (commonly 0.5), New‑AIMD computes a dynamic decrease coefficient based on measured round‑trip time (RTT) and instantaneous queue occupancy. This prevents the congestion window from collapsing too abruptly after a loss event. In the additive increase phase, the algorithm grows the congestion window more aggressively only when the queue is underutilized and RTT remains low; as the queue approaches its limit, the increase step is gradually reduced. These two adaptations aim to smooth the oscillations in window size, thereby reducing queuing delay and improving link utilization.

To evaluate the proposed scheme, the authors employ the NCTUns network simulator. The test topology consists of a single bottleneck router employing a Drop‑Tail active queue management (AQM) policy, with multiple sender‑receiver pairs generating TCP traffic. Experiments vary the number of concurrent flows (5, 10, 20, and 40), while keeping the bottleneck link capacity at 10 Mbps and the baseline RTT at 20 ms. For each scenario, the authors collect average end‑to‑end delay, queue length statistics (mean and standard deviation), and the percentage of time the bottleneck link is fully utilized.

Results show that New‑AIMD consistently outperforms the standard AIMD implementation across all flow counts. Average end‑to‑end delay is reduced by roughly 15 % to 25 % compared with classic AIMD, with the most pronounced gains observed under high contention (40 flows). Queue length variability also drops significantly, indicating a more stable queuing system. Bottleneck link utilization remains above 85 % for New‑AIMD, whereas standard AIMD’s utilization degrades to below 70 % as the number of flows increases. The dynamic decrease factor mitigates the severe window shrinkage that typically follows packet loss, while the adaptive increase prevents the queue from filling too quickly, together yielding fewer packet drops under Drop‑Tail management. Consequently, packet loss rates are lower, and overall throughput is higher.

The authors acknowledge several limitations. The study is confined to simulation; real‑world implementation on actual routers is required to confirm the observed benefits. Only Drop‑Tail AQM is examined; interactions with more sophisticated AQMs such as RED or CoDel remain unexplored. Moreover, the experiments are limited to a 10 Mbps bottleneck and moderate RTT; the scalability of New‑AIMD to high‑speed (≥1 Gbps) or long‑delay (hundreds of ms) networks is not demonstrated. Future work is proposed in three directions: (1) hardware‑level deployment and field testing, (2) extensive evaluation with diverse AQM schemes, and (3) integration of machine‑learning techniques to automatically tune the dynamic coefficients based on real‑time traffic patterns, possibly extending the approach to multipath TCP scenarios. In summary, the paper presents a promising refinement of the AIMD paradigm that achieves lower latency, more stable queues, and higher link utilization, while highlighting the need for broader validation and further optimization.