Enhancing Network Congestion Control: A Comparative Study of Traditional and AI-Enhanced Active Queue Management Techniques

Mohammed Qassim Matrood^1,*, Majid Hamid Ali¹

¹Computer science Department, Tikrit University, Tikrit, 34001, Iraq

Emails: mohammed.q.matrood@st.tu.edu.iq; majid.hamid@tu.edu.iq

Abstract

The issue of multi-access services based on the rapidly expanding Internet affects communication networks and creates congestion problems in buffers, which require effective control. Buffers have previously been managed using simple algorithms such as Droptail (DT), but this method has proven to have many setbacks, such as large queue delays and frequent occurrences of global synchronizations and shutdowns. To overcome these problems, the Active Queue Management (AQM) technique was introduced, including algorithms like Random Early Detection (RED). AQM techniques predict and discharge packets or label them before the buffer reaches its capacity to prevent congestion. In recent work, these algorithms have been enhanced with deep reinforcement learning to achieve improved network performance. This paper intends to present an evaluation of different studies conducted by researchers on congestion control methods. More importantly, it aims to compare the various findings, highlight the prospects of the different methods amid their weaknesses, and discuss future research opportunities within this critical domain of network management.

Keywords: Random Early Detection; RED; Deep Reinforcement Learning; Machine learning; DRL