Buffered Aloha with K-Exponential Backoff -- Part II: Delay Analysis

TL;DR

This paper analyzes delay performance in buffered Aloha networks with K-Exponential Backoff, deriving mean delays, comparing different backoff schemes, and extending the analysis to contention-window models, revealing insights into stability and delay behavior.

Contribution

It provides a comprehensive delay analysis for K-Exponential Backoff in buffered Aloha, including extensions to practical contention-window models and stability insights.

Findings

Higher delay with Geometric Retransmission at low input rates

Severe delay deterioration with Exponential Backoff at high traffic

Retransmission-probability and contention-window models have similar stability regions

Abstract

This paper presents the delay analysis for buffered Aloha networks with K-Exponential Backoff. Mean access delay and mean queueing delay are derived and demonstrated via the examples of Geometric Retransmission (K=1) and Exponential Backoff (K=infinity). The comparison shows that higher delay is incurred with Geometric Retransmission when the aggregate input rate is small, and the delay gap is enlarged as the number of nodes n increases. With a high traffic input rate, however, the delay performance with Exponential Backoff severely deteriorates. The mean queueing delay will be unbounded if the aggregate input rate exceeds 0.3. We also extend the analysis to the contention-window-based backoff model which is widely adopted in practical MAC protocols. It will be revealed that both the retransmission-probability-based and the contention-window-based models exhibit the same stable region and achieve similar queueing performance in most cases, which justifies the intuition that was taken but remained unverified in previous studies: the retransmission-probability-based backoff model can serve as a good approximation of the contention-window-based one.