AD-RED: A new variant of random early detection AQM algorithm

Abstract

Intensive continuing research has been noticed among scholars in the literature with a particular appreciable interest in developing new enhanced variants for the long-standing Random Early Detection (RED) algorithm. Contemporary trends shows that researchers continue to follow a research line thereby exchanging the linear curve needed in RED with nonlinear curves. Several reports have shown that RED’s sole linear function is insufficiently powered for managing rising degrees of traffic congestion in the network. In this paper, Amended Dropping – Random Early Detection (AD-RED), a revised version of RED is presented. AD-RED algorithm consists in combining two nonlinear packet dropping functions: quadratic plus exponential. What’s more, results from ns-3 simulator shows that AD-RED reasonably stabilized and minified the (average) queue size; and obtained a whittled down end-to-end delay when compared with RED itself and another variant of RED. Hence, AD-RED is offered as a fully sufficient replacement for RED’s algorithm implementation in routers.

1. Introduction

Network traffic congestion control remains a persistent and inescapable problem on the Internet due to: (i) increasing magnitude in the number of users deploying several impressive applications and (ii) advancements in communication systems [1–5]. Exclusively, from a technical standpoint, congestion occurs in a network when the proportion of data traffic transmitted increases above the sustainable capacity of the supporting network infrastructure [3,6,7]. Congestion calamitously impacts network performance with regards to problem of large average queue size, increased delay, increased packet loss rate, reduced throughput [3,8–12]. On a more extreme scale, a problem often referred to as congestion collapse can occur when entire communication ceases in the network [13,14].

Two forms of congestion control methods exist: router-based and source-based. The latter method usually implements the Transport Control Protocol (TCP), transport layer protocol. As it turns out, the former method which is implemented in routers effectively enhances network performance and is known to exist in two forms [4,15]: passive queuing and active queuing. Active queue management (AQM) expresses potential to detect incipient congestion and transmits a notification signal to sending devices to whittle down their rates [16–19]. Contrastingly, the use of passive queue management (e.g. DropTail queuing) which inherently relies on First In First Out (FIFO) principle leads to poor network performance as all arriving packets are compulsorily dropped in the event that the buffer is wholly saturated [16,20,21].

One of the novel and earliest AQM algorithm which has been widely investigated in literature is random early detection (RED, as it is called) profoundly initiated by scientists in [22]. RED has been instrumental in providing inspiration to scholars in pursuit of a more effective and robust congestion control algorithms for implementation in routers. RED make do with two thresholds: (i) minimum threshold (denoted $\mathit{T}{\mathit{h}}_{\min }$ ) and (ii) maximum threshold (denoted $\mathit{T}{\mathit{h}}_{\max }$ ). Usually, when a packet arrives the buffer, RED computes the average queue size (denoted ${\mathit{q}}_{\mathrm{ave}}$ ) for congestion indication according to an exponential weighted moving average (EWMA) function Eq. (1): $$\begin{array}{}\text{(1)}& {\mathit{q}}_{\mathrm{ave}}=\mathrm{(}1-\mathit{w}\mathrm{)}\times {\mathit{q}}_{\mathrm{ave}}^{\prime }+\mathrm{(}\mathit{w}\times \mathit{q}\mathrm{)}& \end{array}$$ where ${\mathit{q}}_{\mathrm{ave}}^{\prime }$ refers to ${\mathit{q}}_{\mathrm{ave}}$ computed previously, $\mathit{w}\in [0\mathrm{,}1]$ refers to the averaging weight, q denotes the current queue size.

Afterwards, ${\mathit{q}}_{\mathrm{ave}}$ is compared to the two thresholds. RED accepts an incoming packets when ${\mathit{q}}_{\mathrm{ave}}$ is below $\mathit{T}{\mathit{h}}_{\min }$ , probabilistically drops an incoming packet using a linear approach when $\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }$ , and dismisses (or drop) an arriving packet when ${\mathit{q}}_{\mathrm{ave}}$ towers above $\mathit{T}{\mathit{h}}_{\max }$ . The piecewise function to describe RED’s drop policy is given as follows: $$\begin{array}{}\text{(2)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ \frac{{\mathit{p}}_{\max }\times \mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ Based on Eq. (2), the final drop probability function (denoted ${\mathit{p}}_{\mathit{a}}$ ) is computed as follows: $$\begin{array}{}\text{(3)}& {\mathit{p}}_{\mathit{a}}={\mathit{p}}_{\mathit{b}}\times {\displaystyle \frac{1}{1-\mathrm{(}\text{count }\times {\mathit{p}}_{\mathit{b}}\mathrm{)}}}& \end{array}$$ where ${\mathit{p}}_{\mathit{b}}$ and ${\mathit{p}}_{\mathit{a}}$ refers to the initial; ${\mathit{p}}_{\max }$ parameter denotes the maximum packet dropping probability, and $\text{count}$ represents the number of arrived packets since last packet drop.

More interestingly, since 1993 (referring to the year that RED was introduced), the smooth increasing property of linear dropping function implemented in RED has certainly become an obvious symbol for research, in that, it has been arousing the interest of scholars even on a global scale. One good reason for this is because, in part, it often leads to large delay when congestion in the network becomes severe. Therefore, in recent times, a paradigm shift has been observed in some works with respect to newer AQM algorithm designs. Precisely, new RED-based algorithms that exerts nonlinear drop functions are being initiated as an alternative to the characteristic linear drop function operated in RED, yielding higher network performance.

As rightly pointed out by Kato and colleagues in paper [23], among the class of nonlinear functions, the exponential function is fast gaining recognition as a more powerful dropping function for designing newer RED-based algorithms. In tandem with this line of thought, therefore, the two-fold contribution of this study is as follows: (1) an updated version of RED is proposed named Amended Dropping – Random Early Detection (AD-RED) algorithm. The idea of AD-RED consists in bringing into service the slowly increasing advantage of a quadratic drop function as well as the fast rising benefit of an exponential packet drop function in order to effectively address network congestion, and (2) an investigation of AD-RED’s performance strength in comparison with two (existing) AQM algorithms was done via simulation experiments.

The rest of the paper is outlined along these lines. Section 2 looks at the related works. In Section 3, the principle for the proposed AD-RED algorithm is given. Section 4 offers the simulation configurations, simulation experiments and discussion on results. In closing, the conclusion is offered in Section 5.

2. Related works

In [24], Jafri et al. initiated the Aggressive-RED (AgRED) model. AgRED operates similar but rather chooses to implement sigmoid drop function (with an S-shape) as opposed to the sole linear function exerted in RED. AgRED showcases strength in delay whittling down than RED. In [9], Abdel-Jaber invented the RED-Exponential (RED₋E) model designed to effectively do away with ${\mathit{p}}_{\max }$ ’s parameter overload in RED. RED₋E employs same thresholds: $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ as done by RED, however, an exponential drop function was used in preference to the linear drop function, obtaining a delay performance gain. RED₋E’s dropping function is expressed as follows: $$\begin{array}{}\text{(4)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ \frac{{\mathit{e}}^{{\mathit{q}}_{\mathrm{ave}}}-{\mathit{e}}^{\mathit{T}{\mathit{h}}_{\min }}}{{\mathit{e}}^{\mathit{T}{\mathit{h}}_{\max }}-{\mathit{e}}^{\mathit{T}{\mathit{h}}_{\min }}}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ In [25], a nonlinear RED (NLRED) algorithm was suggested by Zhou et al. NLRED, as done by RED uses same queue threshold parameters, $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ . However, the linear drop function was substituted with a quadratic drop function, achieving an increased throughput. NLRED’s packet drop function is expressed as follows: $$\begin{array}{}\text{(5)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {{\mathit{p}}_{\max }}^{\prime }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^2& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ in which ${{\mathit{p}}_{\max }}^{\prime }=1.5\times {\mathit{p}}_{\max }$

The research in [26] introduced the Gentle RED (GRED) model, another update on RED which utilized three queue threshold parameters: $\mathit{T}{\mathit{h}}_{\min }$ , $\mathit{T}{\mathit{h}}_{\max }$ , and $2\times \mathit{T}{\mathit{h}}_{\max }$ . In GRED, two linear packet drop functions were used. This was done in order to keep at minimum the aggressive feature of the linear packet drop function of RED and is therefore expressed as follows: $$\begin{array}{}\text{(6)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ {\mathit{p}}_{\max }+\mathrm{(}1-{\mathit{p}}_{\max }\mathrm{)}\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\max }}{\mathit{T}{\mathit{h}}_{\max }}\mathrm{)}& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}2\times \mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }2\times \mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ Scholars in [27] invented MRED model which resembles GRED model in that it utilizes three threshold parameters namely, $\mathit{T}{\mathit{h}}_{\min }$ , $\mathit{T}{\mathit{h}}_{\max }$ , and $2\times \mathit{T}{\mathit{h}}_{\max }$ . But unlike GRED, the MRED model obtained a higher throughput performance due to its usage of the conjunction of two different dropping functions- quadratic as well as linear, expressed as follows: $$\begin{array}{}\text{(7)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }\mathrm{(}\frac{{{\mathit{q}}_{\mathrm{ave}}}^2-{\mathit{T}{\mathit{h}}_{\min }}^2}{{\mathit{T}{\mathit{h}}_{\max }}^2-{\mathit{T}{\mathit{h}}_{\min }}^2}\mathrm{)}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ {\mathit{p}}_{\max }+\frac{\mathrm{(}1-{\mathit{p}}_{\max }\mathrm{)}\times \mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\max }\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }}& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}2\times \mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }2\times \mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ The study in [28] explicitly applies one (linear) as well as another (exponential) dropping functions for extending RED algorithm in order that a lower average queue size would be achieved. The improved algorithm was named modified dropping-RED (abbreviated MD-RED). The model operates within the boundary set by RED (i.e., between $\mathit{m}\mathit{i}\mathit{n}\mathit{T}\mathit{R}$ and $\mathit{m}\mathit{a}\mathit{x}\mathit{T}\mathit{R}$ ). In [23], Kato et al. initiated RED with nonlinearity which resembles NLRED model, but unlike NLRED, the modified model simply substituted the squaring function in NLRED with an exponential function (Eq. (8)) in order to achieve a higher throughput benefit. $$\begin{array}{}\text{(8)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}{\mathit{e}}^{\mathit{\gamma }[\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}-1]}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ in which the parameter gamma (γ) is varied in order to achieve a suitable bending degree for the exponential curve.

Based on RED mechanism, the authors in [29] proposed RED-Exponential and RED-Linear analytical models which uses instantaneous queue length for congestion detector. Authors in [30] suggested Quadratic RED (QRED) algorithm as an amendment for NLRED. QRED utilized the $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ queue thresholds while bringing to play a quadratic packet drop function given as follows: $$\begin{array}{}\text{(9)}& {\mathit{p}}_{\mathit{b}}={\mathrm{\left(}{\displaystyle \frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{K}-\mathit{T}{\mathit{h}}_{\min }}}\mathrm{\right)}}^2& \end{array}$$ or $$\begin{array}{}\text{(10)}& {\mathit{p}}_{\mathit{b}}=1-{\mathrm{\left(}{\displaystyle \frac{\mathit{K}-{\mathit{q}}_{\mathrm{ave}}}{\mathit{K}-\mathit{T}{\mathit{h}}_{\min }}}\mathrm{\right)}}^2& \end{array}$$ in which K denotes the size of buffer size.

In [4], a Beta distribution function for computing the packet drop probability was called into play when ${\mathit{q}}_{\mathrm{ave}}$ is bounded between $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ threshold values in order to stabilize the queue. In [31], the Congestion Control RED (CoCo-RED) model which aimed at keeping to a minimum the number of dropped packets was initiated. Using CoCo-RED, a conjunction of two drop actions were deployed. Firstly, a linear packet drop function is exerted between the bounds of $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ threshold parameters. Secondly, an exponential packet drop function is utilized between the bounds of $\mathit{T}{\mathit{h}}_{\max }$ and the buffer size (represented by K). The modified packet dropping probability function is expressed as: $$\begin{array}{}\text{(11)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ \frac{{\mathit{p}}_{\max }\times \mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ \mathit{a}{\mathit{b}}^{{\mathit{q}}_{\mathrm{ave}}}& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{K}\mathrm{,}\end{array}\right.& \end{array}$$ in which $$\begin{array}{}\text{(12)}& \mathit{a}={\displaystyle \frac{{\mathit{p}}_{\max }}{{\mathrm{(}{\mathit{e}}^{\frac{\ln \mathrm{(}1\mathrm{/}{\mathit{p}}_{\max }\mathrm{)}}{\mathit{K}-\mathit{T}{\mathit{h}}_{\max }}}\mathrm{)}}^{\mathit{T}{\mathit{h}}_{\max }}}}& \end{array}$$ and $$\begin{array}{}\text{(13)}& \mathit{b}={\mathit{e}}^{\frac{\ln \mathrm{(}1\mathrm{/}{\mathit{p}}_{\max }\mathrm{)}}{\mathit{K}-\mathit{T}{\mathit{h}}_{\max }}}& \end{array}$$ In Double Slope (DS)-RED [32], the bounds between $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ thresholds is divided equally (involving a mid-point set as ( $\mathit{T}{\mathit{h}}_{\min }+\mathit{T}{\mathit{h}}_{\max }\mathrm{)}\mathrm{/}2$ ), each part utilizing a linear function, however, with different slopes. By so doing, DSRED cut back on the aggressive characteristics of RED and increases its throughput. Domańska et al. in [33], also utilized the $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ threshold positions of RED. However, a cubic packet dropping probability function was brought into action between the two positions. In [34], Smart RED (SmRED) model employs three threshold parameters: $\mathit{T}{\mathit{h}}_{\min }$ , $\mathit{T}{\mathit{h}}_{\text{target}}$ (set as $\mathit{T}{\mathit{h}}_{\min }+\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}\mathrm{/}2$ ), and $\mathit{T}{\mathit{h}}_{\max }$ . A quadratic function is used in conjunction with a square root packet drop function. Dropping function in SmRED can be expressed as follows: $$\begin{array}{}\text{(14)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^2& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\text{target}}\mathrm{,}\\ {\mathit{p}}_{\max }\sqrt{\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}}& \text{when }\mathit{T}{\mathit{h}}_{\text{target}}⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ The authors in [35] perceived the distinctive linear dropping function in RED as low-powered for effective congestion control. Therefore, they suggested an extension to RED, named RED-Quadratic linear (RED-QL) which utilizes both quadratic and linear functions. RED-QL exhibited gain in delay whittling down. In [15], the authors initiates an improved RED algorithm that modifies the drop probability used between $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ threshold parameters. Dropping function used can be expressed as follows: $$\begin{array}{}\text{(15)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ 1-\mathrm{(}{\mathit{p}}_1\times \frac{-\log \mathrm{(}{\mathit{p}}_1\mathrm{)}}{\text{count}+1}\mathrm{)}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ in which $$\begin{array}{}\text{(16)}& {\mathit{p}}_1={\mathit{p}}_{\max }\times \mathrm{\left(}{\displaystyle \frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}}\mathrm{\right)}& \end{array}$$ Based on CoCo-RED algorithm, the authors in [36] chose to dynamically adjust the $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ thresholds. The improved scheme was called Enhanced CoCo-RED (EnCoCo-RED). An enhanced SmRED algorithm named SmRED-i was initiated by authors in [37]. SmRED-i utilizes the three queue thresholds of SmRED. However, a parameter $\mathit{i}=2\mathrm{,}3\mathrm{,}4\mathrm{,}5\mathrm{,}\dots$ was brought to bear. SmRED-i’s packet drop function is given as follows: $$\begin{array}{}\text{(17)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^{\mathit{i}}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\text{target}}\mathrm{,}\\ {\mathit{p}}_{\max }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^{1\mathrm{/}\mathit{i}}& \text{when }\mathit{T}{\mathit{h}}_{\text{target}}⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ The paper in [38] shares the opinion that a typical linear function used by RED is underpowered. The authors suggested a model which uses two linear functions, it was named Double Linear-RED (DL-RED). It displays considerable strength in delay reduction. In [39], the Adaptive GRED (AGRED) was initiated similar to GRED. However, the modification lies where ${\mathit{q}}_{\mathrm{ave}}$ falls in between $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ parameters. AGRED’s packet drop function is given as follows: $$\begin{array}{}\text{(18)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ {\mathit{p}}_{\max }+\frac{\mathrm{(}1-{\mathit{p}}_{\max }\mathrm{)}\times \mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\max }\mathrm{)}}{2\times \mathit{T}{\mathit{h}}_{\max }}& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}2\times \mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }2\times \mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ The author in [40] initiated another improved version of RED named RED-Linear Exponential (RED-LE). This algorithm cut in half the portion bounded by $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ thresholds of RED utilizing linear as well as exponential drop functions. RED-LE succeeded in keeping at minimum the average queue size. $$\begin{array}{}\text{(19)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ 2{\mathit{p}}_{\max }\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\text{target}}\mathrm{,}\\ {\mathit{e}}^{\log \mathrm{(}{\mathit{p}}_{\max }\mathrm{)}[\frac{2\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-{\mathit{q}}_{\mathrm{ave}}\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}]}& \text{when }\mathit{T}{\mathit{h}}_{\text{target}}⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ in which $$\begin{array}{}\text{(20)}& \mathit{T}{\mathit{h}}_{\text{target}}={\displaystyle \frac{\mathit{T}{\mathit{h}}_{\max }+\mathit{T}{\mathit{h}}_{\min }}{2}}& \end{array}$$ In Three-Section RED (TRED) [41] model, three queue thresholds parameters are utilized: $\mathit{T}{\mathit{h}}_{\min }$ , Δ (set as $\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}\mathrm{/}3$ ), and $\mathit{T}{\mathit{h}}_{\max }$ . TRED make do with a nonlinear (i.e., cubic), linear, plus an extra nonlinear (i.e., cubic) packet drop functions in the first, second, and third segments respectively. This was done in order to attain a trade-off between delay as well as throughput under varying traffic loads. TRED’s packet drop function is given as follows: $$\begin{array}{}\text{(21)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ 9{\mathit{p}}_{\max }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^3& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }+\mathrm{\Delta }\mathrm{,}\\ \frac{{\mathit{p}}_{\max }\times \mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}{\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}& \text{when }\mathit{T}{\mathit{h}}_{\min }+\mathrm{\Delta }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }+2\mathrm{\Delta }\mathrm{,}\\ {\mathit{p}}_{\max }+9{\mathit{p}}_{\max }{\mathrm{(}\frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\max }}{\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}^3& \text{when }\mathit{T}{\mathit{h}}_{\min }+2\mathrm{\Delta }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ To further increase the throughput performance of RED, [42] suggested an improved adaptive RED. Based on RED, the new variant utilizes a cubic drop function when ${\mathit{q}}_{\mathrm{ave}}$ increases above $\mathit{T}{\mathit{h}}_{\min }$ but lesser than $\mathit{T}{\mathit{h}}_{\max }$ . Mathematically, the piecewise function to describe improved adaptive RED drop policy is given as follows: $$\begin{array}{}\text{(22)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0\mathrm{,}& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ {\mathit{p}}_{\max }[\frac{3{\mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}^2}{{\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}^2}-\frac{2{\mathrm{(}{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}^3}{{\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}^3}]\mathrm{,}& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1\mathrm{,}& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\end{array}\right.& \end{array}$$

In light of the review made thus far, it is observed that newer AQM algorithms continue to share the believe that utilizing a nonlinear function in the piecewise function for describing the packet drop probability function of RED will go a long way in increasing its performance rather than adopting just one linear function. This claim is also supported by [4,30,34]. Moreover, obtaining average queue size lowering is an important goal of AQM in which existing algorithms could not meet. The substantial innovation of this study is two-fold. Firstly, a new proposal is presented called Amended Dropping – Random Early Detection (AD-RED) with reference to RED algorithm. Concretely, AD-RED aimed at maintaining a network’s average queue size lowering as a means for ensuring a minified delay network performance. AD-RED achieves this by dividing the queue into four sections and put to use the composition of double nonlinear drop functions.

3. Amended Dropping - Random Early Detection (AD-RED) algorithm

The proposed Amended Dropping – Random Early Detection (AD-RED) AQM algorithm is designed based on RED. Essentially, in AD-RED, the dropping function of RED is revised consisting of two nonlinear dropping functions as against the sole linear dropping function. Concretely, the mathematical model (describing the working procedure) for AD-RED is expressed using a piecewise function as follows: $$\begin{array}{}\text{(23)}& {\mathit{p}}_{\mathit{b}}=\left\{\begin{array}{ll}0& \text{when }{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }\mathrm{,}\\ 9{\mathit{p}}_{\max }{\mathrm{(}\frac{\mathrm{ave}-\mathit{T}{\mathit{h}}_{\min }}{2\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-2\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}\mathrm{)}}^2& \text{when }\mathit{T}{\mathit{h}}_{\min }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\text{target}}\mathrm{,}\\ {\mathit{e}}^{\log \mathrm{(}{\mathit{p}}_{\max }\mathrm{)}\times \mathrm{(}\frac{3\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-{\mathit{q}}_{\mathrm{ave}}\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }+\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}& \text{when }\mathit{T}{\mathit{h}}_{\text{target}}⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }\mathrm{,}\\ 1& \text{when }\mathit{T}{\mathit{h}}_{\max }⩽{\mathit{q}}_{\mathrm{ave}}\mathrm{,}\end{array}\right.& \end{array}$$ As could be observed from Eq. (23), AD-RED utilizes an extra queue threshold, $\mathit{T}{\mathit{h}}_{\text{target}}$ for the sake of cutting the queue part of RED bounded by $\mathit{T}{\mathit{h}}_{\min }$ and $\mathit{T}{\mathit{h}}_{\max }$ positions into two unequal portions. The value for $\mathit{T}{\mathit{h}}_{\text{target}}$ (is set based on Eq. (24)). This way, two sub-intervals of unequal lengths is achieved. One (i.e., the shorter) exists between $\mathit{T}{\mathit{h}}_{\min }$ and Target (account for light and moderate traffic loads), while the other exist between Target and $\mathit{T}{\mathit{h}}_{\max }$ (account for heavy traffic load). $$\begin{array}{}\text{(24)}& \mathit{T}{\mathit{h}}_{\text{target}}=2\mathrm{\left(}{\displaystyle \frac{\mathit{T}{\mathit{h}}_{\max }+\mathit{T}{\mathit{h}}_{\min }}{3}}\mathrm{\right)}-\mathit{T}{\mathit{h}}_{\min }& \end{array}$$ At light as well as moderate traffic load scenarios whereby congestion is rated as not severe yet, AD-RED model cut back on the aggressiveness of RED model. Such that, packets are dropped by bringing into action a quadratic function – Eq. (25). This account for cases when ${\mathit{q}}_{\mathrm{ave}}⩾\mathit{T}{\mathit{h}}_{\min }$ as well as ${\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\text{target}}$ $$\begin{array}{}\text{(25)}& {\mathit{p}}_{\mathit{b}}=9{\mathit{p}}_{\max }{\mathrm{\left(}{\displaystyle \frac{{\mathit{q}}_{\mathrm{ave}}-\mathit{T}{\mathit{h}}_{\min }}{2\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-2\mathit{T}{\mathit{h}}_{\min }\mathrm{)}}}\mathrm{\right)}}^2& \end{array}$$ At heavy network traffic load scenarios whereby congestion is seems severe, AD-RED quickly increases the aggressiveness of RED model. This way, packets dropping is achieved by calling into play an exponential function – Eq. (26). This account for cases when ${\mathit{q}}_{\mathrm{ave}}⩾\mathit{T}{\mathit{h}}_{\text{target}}$ as well as ${\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\max }$ $$\begin{array}{}\text{(26)}& {\mathit{p}}_{\mathit{b}}={\mathit{e}}^{\log \mathrm{(}{\mathit{p}}_{\max }\mathrm{)}\times \mathrm{(}\frac{3\mathrm{(}\mathit{T}{\mathit{h}}_{\max }-{\mathit{q}}_{\mathrm{ave}}\mathrm{)}}{\mathit{T}{\mathit{h}}_{\max }+\mathit{T}{\mathit{h}}_{\min }}\mathrm{)}}& \end{array}$$ Furthermore, when ${\mathit{q}}_{\mathrm{ave}}\mathrm{<}\mathit{T}{\mathit{h}}_{\min }$ , incoming packets are admitted (that is, ${\mathit{p}}_{\mathit{b}}=0$ ) while all incoming packets are discarded when ${\mathit{q}}_{\mathrm{ave}}⩾\mathit{T}{\mathit{h}}_{\max }$ (that is, ${\mathit{p}}_{\mathit{b}}=1$ ), as done by RED. Therefore, in Algorithm 1, the pseudocode illustrating the major steps involved in AD-RED is presented. Thence, commensurate with the above-mentioned idea, by configuring $\mathit{T}{\mathit{h}}_{\min }$ to 30 packets, ${\mathit{p}}_{\max }$ to 0.1 (as suggested by scholars in [22]), and $\mathit{T}{\mathit{h}}_{\max }$ to $3\times \mathit{T}{\mathit{h}}_{\min }=90$ packets (also as recommended by scientists in [22]), the proposed AD-RED dropping function is visualized in Fig. 1.

Algorithm 1

Pseudocode for AD-RED

Fig. 1.

AD-RED’s dropping function.

4. Simulation experiment and results

In this section, the proposed AD-RED algorithm is implemented in ns-3 [43]. Furthermore, its performance is assessed against two AQM algorithms: TRED as well as RED.

4.1. Simulation configuration

The network topology considered for simulation is shown in Fig. 2. The topology has N number of sources (which is varied), two routers: A and B, and one sink (i.e., destination end-system). The three AQM algorithms being evaluated were deployed at router A (in a sequential arrangement). Each source end-system is connected to router A using a point-to-point link of 100- $\mathit{Mbit}\mathrm{/}\mathit{s}$ with 10- $\mathit{ms}$ propagation delay time. Router A is then connected to router B using a bottleneck point-to-point link of 10- $\mathit{Mbit}\mathrm{/}\mathit{s}$ with 100- $\mathit{ms}$ propagation delay time. Similarly, router B is connected to the sink using a point-to-point link of 100- $\mathit{Mbit}\mathrm{/}\mathit{s}$ with 10- $\mathit{ms}$ propagation delay time. TCP New-Reno was implemented. The default packet size is 1000 $\mathit{bytes}$ long. Buffer size is set to 250 packets. Simulation duration is set to 100 $\sec$ while $\mathit{w}=0.002$ (as recommended by scientists in [22]). The number of sender nodes is varied as a way of offer varying levels of congestion on the network which is in tandem with the suggestions made in RFC 7928 [44].

Fig. 2.

Network topology simulated.

4.2. Experiment 1: Light load

In this simulation experiment, 5 TCP sources are considered which account for light load scenario. In Fig. 3(a), the average queue size is depicted. It can be observed that AD-RED keeps down the average queue size better than RED and even much smaller relative to both TRED and RED. This result is also confirmed in Table 1. Figure 3(b) shows the delay performance of the trio algorithms. It can be observed that AD-RED algorithm exhibits a significant improvement over RED and TRED. The results depicted in Fig. 3(c) shows the throughput performance for the three algorithms. Statistical result supporting this claim is provided in Table 2. TRED obtained the highest throughput value while AD-RED got the least value even as presented in Table 3.

Fig. 3.

Performance comparison for AD-RED, TRED, and RED under light traffic condition.

Table 1

Average queue size comparison – light load

Statistical analysis	Average queue size, Packets
	AD-RED	TRED	RED
Mean	7.7802	12.6164	8.9880
Standard deviation	2.2932	2.8189	3.1508
Maximum	28.1297	34.1103	31.3134

Table 2

Delay comparison – light load

Statistical analysis	Delay, ms
	AD-RED	TRED	RED
Mean	1.4769	1.5507	1.4924
Standard deviation	0.1014	0.1098	0.1051
Maximum	1.6970	1.7162	1.7161

Table 3

Throughput comparison – light load

Statistical analysis	Throughput, Mbps
	AD-RED	TRED	RED
Mean	9.4392	9.5590	9.4764
Standard deviation	1.1330	1.1476	1.1329
Maximum	9.8199	9.9225	9.8772

4.3. Experiment 2: Moderate load

In this simulation experiment, 50 TCP sources are considered which account for moderate load scenario. In Fig. 4(a), AD-RED again succeeded over both TRED as well as RED with regards to achieving a good cut back on the average queue size. Results depicted in Fig. 4(b) confirms that AD-RED is more effective when compared with TRED as well as RED algorithms in obtaining a reduced delay even when the traffic load is moderate. Figure 4(c) demonstrates the throughput performance for the three algorithms. RED performed better in contrast to TRED and AD-RED algorithms. However, TRED outperformed AD-RED. Tables 4–6 presents the statistical results for average queue size, delay, as well as throughput respectively.

Fig. 4.

Performance comparison for AD-RED, TRED, and RED under moderate traffic condition.

Table 4

Average queue size comparison – moderate load

Statistical analysis	Average queue size, Packets
	AD-RED	TRED	RED
Mean	22.5339	35.8891	36.4716
Standard deviation	2.8146	2.9878	2.9733
Maximum	49.3588	58.7594	58.7507

Table 5

Delay comparison – moderate load

Statistical analysis	Delay, ms
	AD-RED	TRED	RED
Mean	14.9898	15.9879	16.1475
Standard deviation	1.9288	2.1070	2.1283
Maximum	15.7995	16.7475	16.9302

Table 6

Throughput comparison – moderate load

Statistical analysis	Throughput, Mbps
	AD-RED	TRED	RED
Mean	9.3360	9.7542	9.8081
Standard deviation	0.7685	0.8299	0.8326
Maximum	9.4798	9.9261	9.9707

4.4. Experiment 3: Heavy load

In this simulation experiment, 100 TCP sources which account for heavy load scenario are considered. In Fig. 5(a), the average queue size for AD-RED, TRED, as well as RED is compared. It can be clearly seen that the proposed AD-RED model exhibit superior advantage over TRED and RED with respect to keeping at minimum the average queue size. Although, TRED achieved a smaller average queue size compared with RED. As Fig. 5(c) illustrates, it is evident that the performance of AD-RED algorithm is superior to both TRED and RED with respect to cutting back on delay even at a heavy traffic load. The throughput performance of AD-RED, TRED, as well as RED is depicted in Fig. 5(b). RED obtained the highest value while TRED outperfromed AD-RED. Tables 7–9 provides further statistical results regarding the performance metrics.

Fig. 5.

Performance comparison for AD-RED, TRED, and RED under heavy traffic condition.

Table 7

Average queue size comparison – heavy load

Statistical analysis	Average queue size, Pkt
	AD-RED	TRED	RED
Mean	27.8964	49.0254	51.3717
Standard deviation	2.3291	3.6993	4.0873
Maximum	44.0389	51.6655	52.9229

Table 8

Delay comparison – heavy load

Statistical analysis	Delay, ms
	AD-RED	TRED	RED
Mean	28.7142	32.5021	32.7001
Standard deviation	6.6832	7.4041	7.4285
Maximum	31.9644	36.1611	34.4336

Table 9

Throughput comparison – heavy load

Statistical analysis	Throughput, Mbps
	AD-RED	TRED	RED
Mean	9.2816	9.8558	9.8509
Standard deviation	0.7415	0.8090	0.8096
Maximum	9.4248	10.0248	10.0214

5. Conclusions

In recent past years, congestion control at the router has been an important subject for discussion among researchers and scientists in computer networking field. Developing an active queue management (AQM) algorithm that reasonably modifies the dropping function of Random Early Detection (RED) has consistently remained a major research hotspot, catching the attention of researchers all around the world. Therefore, this paper has presented a revised AQM algorithm named Amended Dropping – RED (AD-RED). Simulation evaluations in ns3 simulator asserts that the dropping functions used in AD-RED are indeed a promising combination for improving RED’s performance. AD-RED exhibits good advantage in improving delay performance when compared with RED as well as one other RED-based AQM algorithms. In near future work, AD-RED will be implemented in cellular networks.

Conflict of interest

The author has no conflict of interest to report.