On Effectiveness of Backlog Bounds Using Stochastic Network Calculus in 802.11

TL;DR

This paper investigates the effectiveness of stochastic network calculus in bounding backlogs in 802.11 wireless LANs, deriving theoretical bounds, comparing with simulations, and proposing techniques for improvement.

Contribution

It models 802.11 nodes using stochastic network calculus, derives backlog bounds, and explores methods to tighten these bounds for practical applications.

Findings

Derived stochastic service curve for 802.11 nodes.

Found that existing bounds are loose compared to simulations.

Martingale and independent analysis techniques improve bounds.

Abstract

Network calculus is a powerful methodology of characterizing queueing processes and has wide applications, but few works on applying it to 802.11 by far. In this paper, we take one of the first steps to analyze the backlog bounds of an 802.11 wireless LAN using stochastic network calculus. In particular, we want to address its effectiveness on bounding backlogs. We model a wireless node as a single server with impairment service based on two best-known models in stochastic network calculus: Jiang's and Ciucu's. Interestingly, we find that the two models can derive equivalent stochastic service curves and backlog bounds in our studied case. We prove that the network-calculus bounds imply stable backlogs as long as the average rate of traffic arrival is less than that of service, indicating the theoretical effectiveness of stochastic network calculus in bounding backlogs. From A. Kumar's 802.11 model, we derive the concrete stochastic service curve of an 802.11 node and its backlog bounds. We compare the derived bounds with ns-2 simulations and find that the former are very loose and we discuss the reasons. And we show that the martingale and independent case analysis techniques can improve the bounds significantly. Our work offers a good reference to applying stochastic network calculus to practical scenarios.