[Paper Review] Distributed Broadcasting in Wireless Networks under the SINR Model

Tomasz Jurdziński, Dariusz R. Kowalski+2 more|arXiv (Cornell University)|Jul 28, 2012

Mobile Ad Hoc Networks19 references3 citations

TL;DR

This paper presents the first provably efficient and scalable distributed broadcast algorithms for wireless networks under the SINR model. It introduces two randomized protocols: one using local density estimation to achieve broadcast in $O(D + "log(1/ delta))$ rounds with probability $1- delta$, and another without density knowledge that runs in $O((D + delta)) delta)$ rounds, both significantly outperforming traditional exponential backoff in simulations.

ABSTRACT

In the advent of large-scale multi-hop wireless technologies, such as MANET, VANET, iThings, it is of utmost importance to devise efficient distributed protocols to maintain network architecture and provide basic communication tools. One of such fundamental communication tasks is broadcast, also known as a 1-to-all communication. We propose several new efficient distributed algorithms and evaluate their time performance both theoretically and by simulations. First randomized algorithm accomplishes broadcast in O(D+log(1/d)) rounds with probability at least 1-d on any uniform-power network of n nodes and diameter D, when equipped with local estimate of network density. Additionally, we evaluate average performance of this protocols by simulations on two classes of generated networks - uniform and social - and compare the results with performance of exponential backoff heuristic. Ours is the first provably efficient and well-scalable distributed solution for the (global) broadcast task. The second randomized protocol developed in this paper does not rely on the estimate of local density, and achieves only slightly higher time performance O((D+log(1/d))log n). Finally, we provide a deterministic algorithm achieving similar time O(D log^2 n), supported by theoretical analysis.

Motivation & Objective

• To design efficient, scalable distributed broadcast protocols for large-scale ad-hoc wireless networks under the SINR model.
• To address the challenge of achieving low-latency broadcast without global knowledge, leveraging only local information or no extra assumptions.
• To provide theoretical guarantees on time complexity while maintaining practical feasibility through simulation evaluation.
• To overcome limitations of prior work by ensuring correctness and efficiency for any connected network with $ eq < 1/2$.
• To introduce novel distributed leader election techniques suitable for use in broader communication tasks.

Proposed method

• Proposes a randomized distributed algorithm using a grid-based spatial decomposition into boxes of size $ delta)$ rounds, with probability $1- delta$.
• Employs a multi-layered protocol with three phases per round: leader election, message transmission, and interference control.
• Uses a loop-based transmission probability doubling strategy to balance exploration and collision avoidance.
• Applies a $(1- eq/2)$-net construction over a grid to ensure coverage and connectivity across the network.
• Introduces a leader election mechanism within each box that succeeds with probability at least $1/18$ per round.
• Employs a probabilistic analysis to bound the failure probability of message propagation across the network's diameter $D$.

Experimental results

Research questions

• RQ1 Can a distributed broadcast protocol be designed under the SINR model that achieves near-optimal time complexity with high probability?
• RQ2 How can efficient broadcast be achieved without relying on global network parameters like total node count or global density?
• RQ3 What is the performance gap between provably correct algorithms and heuristic approaches like exponential backoff in realistic network topologies?
• RQ4 Can novel distributed leader election techniques be used to build scalable and efficient broadcast protocols?
• RQ5 What theoretical time bounds can be proven for distributed broadcast in the SINR model under minimal assumptions?

Key findings

• The first protocol achieves broadcast in $O(D + delta)$ rounds with probability $1- delta$, using local density estimates.
• The second protocol, without density knowledge, achieves $O((D + delta)) delta)$ rounds, with only a logarithmic factor penalty.
• Simulations on uniform and social network topologies show the proposed protocols outperform exponential backoff in average performance.
• The probability of successful leader election within any box is at least $1/18$ per round, enabling reliable message forwarding.
• The total number of rounds is bounded by $O( delta)$ rounds, with high probability, using $T = O(D + delta)$ repetitions of the main loop.
• The algorithm ensures that each box in the network's $(1- eq/2)$-net receives and forwards the message, guaranteeing global broadcast.