Throughput-Optimal Random Access: A Queueing-Theoretical Analysis for Learning-Based Access Design

TL;DR

This work tackles throughput optimization in random access networks by marrying learning-based strategies with queueing theory. It introduces two MAB-based schemes, MTOA-L (local rewards) and MTOA-G (global rewards), both capable of achieving maximum network throughput, but exhibiting different short-term fairness profiles. A unified queueing-theoretical framework is developed to analyze and optimize the throughput-fairness tradeoff, enabling principled parameter tuning of the learning schemes. The results show that MTOA-G significantly outperforms MTOA-L in large networks, delivering near-maximal throughput under tight fairness constraints and providing a practical blueprint for designing learning-based access protocols with theoretical performance guarantees.

Abstract

Random access networks have long been observed to suffer from low throughput if nodes' access strategy is not properly designed. To improve the throughput performance, learning-based approaches, with which each node learns from the observations and experience to determine its own access strategy, have shown immense potential, but are often designed empirically due to the lack of theoretical guidance. As we will demonstrate in this paper, the queueing-theoretical analysis can be leveraged as a powerful tool for optimal design of learning-based access. Specifically, based on a Multi-Armed-Bandit (MAB) framework, two random access schemes, MTOA-L with local rewards and MTOA-G with global rewards, are proposed for throughput optimization. Though both can achieve the maximum throughput of 1, they have different short-term fairness performance. Through identifying the access strategies learned via MTOA-L and MTOA-G and feeding them into the proposed unified queueing-theoretical framework, the throughput-fairness tradeoff of each is characterized and optimized by properly tuning the key parameters. The comparison of the optimal tradeoffs shows that MTOA-G is much superior to MTOA-L especially when the number of nodes is large.

Figures (7)

• Figure 1: Simulated network throughput $\hat{\lambda}_{out}$ and short-term fairness index $J_T$ with MTOA-L versus the number of null actions $L$. $n=100$. $T=10^7$ slots. (a) $Q_{th}=0$. $\alpha\in\{0.1,0.5,0.9,1\}$. (b)-(c) $Q_{th}\in\{0.001,0.01,0.1,1\}$. $\alpha\in\{0.5,0.9\}$.
• Figure 2: Simulated network throughput $\hat{\lambda}_{out}$ and short-term fairness index $J_T$ with MTOA-G versus the number of null actions $L$. $n=100$. $T=10^7$ slots. (a) $M=\infty$. $\alpha\in\{0.1,0.5,0.9,1\}$. (b)-(c) $\alpha=0.9$. $M\in\{5,5\times 10^2,5\times 10^4\}$.
• Figure 3: Embedded Markov chain $\mathbf{X}$ of the state transition process of an individual HOL batch in the Aloha network.
• Figure 4: Network throughput $\hat{\lambda}_{out}$ versus short-term fairness index $J_T$. $n=100$. $T=10^7$ slots. $K=n_C$. $n_C\in\{0,1,2,3,4\}$. (a) Connection-free Aloha. $q_{n_C}\in[10^{-10}, 10^{-2}]$. (b) Connection-based Aloha. $q_{n_C}\in[10^{-10}, 10^{-1}]$ and $M\in[1,10^6]$ are jointly optimized.
• Figure 5: Analytical and simulated network throughput $\hat{\lambda}_{out}$ and short-term fairness index $J_T$ for MTOA-L. $n=100$. $T=10^7$ slots. $\alpha\in\{0.8,0.9,1\}$. (a) $\hat{\lambda}_{out}$ and $J_T$ versus the number of null actions $L$. $Q_{th}=1$. (b) $\hat{\lambda}_{out}$ versus $J_T$. $Q_{th}\in\{0.005,0.05,0.1\}$. $L\in[10^2,10^6]$.