Elsevier

Ad Hoc Networks

Volume 7, Issue 7, September 2009, Pages 1271-1284
Ad Hoc Networks

Encounter-based message propagation in mobile ad-hoc networks

https://doi.org/10.1016/j.adhoc.2008.12.003Get rights and content

Abstract

A family of message propagation protocols for highly mobile ad-hoc networks is defined, and is studied analytically and by simulation. The coverage of a message (the fraction of nodes that receive it), can be made arbitrarily close to 1, at a moderate cost of extra message traffic. Under certain simplifying assumptions, it is shown that a high coverage is achieved by making a total of O(n ln n) broadcasts, where n is the number of nodes, and the time to propagate a message is O(ln n). Mechanisms for making the protocols more efficient, by reducing the number of redundant transmissions without affecting the achievable coverage, are presented and evaluated. The generalisation to multiple broadcasts proceeding in parallel is derived. Finally, an attempt to validate the proposals in the context of a real-life mobile ad-hoc network is described. Results of this validation exercise and simulations point out that the proposed protocols can cost-effectively achieve high coverage in networks of varying degrees of node densities and mobilities.

We acknowledge the financial support from the UK EPSRC Projects “Networked Computing in Inter-organisation Settings” (EP/D037743/1) and “Protocols for Ad-hoc Collaborative Environments” (GR/S02082/01).

Introduction

Recent advances in the technologies of mobile devices and wireless communication have given rise to an increasingly popular form of networking, called mobile ad-hoc networking. A mobile md-hoc network (MANET) consists of small, versatile and powerful mobile computing devices (nodes). It is typically formed at short notice and does not make use of any fixed networking infrastructure. A distinguishing feature of a MANET is that the nodes are not just the sources of message traffic but also engage in forwarding messages to final destinations; given that the nodes can be highly mobile, a MANET is a dynamic network characterised by frequent and hard-to-predict topological changes.

An application of a mobile network usually involves user collaboration towards achieving a common goal, in situations where access to base stations is unavailable or unreliable (e.g., command and control or disaster relief). The success of such collaborative undertakings depends to a large extent on the provision of reliable, fast and economic multicast. That is, a message originating at any node should reach all other nodes within a reasonably short period of time and without consuming much of the network resources. Unfortunately, both the nature of the devices and their mobility imply that these are conflicting objectives whose achievement cannot normally be guaranteed. One is therefore obliged to trade off coverage (i.e., the fraction of nodes that receive a message) against message traffic and delays.

Our objective is to devise and evaluate message propagation protocols which achieve very high coverage without paying too high a price in terms of propagation delays. We also introduce mechanisms aimed at reducing the price paid in terms of message traffic.

Existing work in this area has concentrated on different aspects of the above trade-offs, with varying degree of success depending on the nature of the network. For example, several protocols (see [9], [25], [10], [3]) aim to minimise the number of broadcasts by using state information about the network topology. When the degree of mobility is low, these protocols perform well, but when it is high, the network state information can become out-of-date quickly and the coverage achieved can be poor [23], [28].

A topology-independent and stateless protocol that seems to work better in highly mobile networks is ‘flooding’; see Ho et al. [14]. In its simplest form, every node broadcasts every message once, either immediately upon receipt or after a random interval. The coverage achieved by flooding depends considerably on the mobility pattern and on the ‘density’ of nodes (usually defined as the average number of nodes within a disc of radius equal to the wireless range). When the density is high, the coverage is potentially high, but flooding causes ‘broadcast storms’, with their attendant problems of wasted network resources and possibility of collisions. To counter these problems, Ni et al. [21] have proposed and studied a number of optimisations, some of which have applications in our protocols.

On the other hand, low density or particular mobility patterns render the network liable to ‘partitioning’ (Hahner et al., [12]). For such networks the flooding coverage tends to be poor (Khelil et al., [17], Obraczka et al., [22]).

Modifications of the flooding protocol, such as ‘Hypergossiping’ [17], [16] and ‘Adaptive-Flooding’ [27] aim to improve coverage by allowing nodes to make more than one broadcast of a message during the life-time of that message. However, the efficacy of those refinements depends heavily on being able to select the life-time duration appropriate to the node speed. The speed is not easy to determine without providing the nodes with special equipment, and even when determined, it is not clear what is the best choice for fixing the life-time duration. Appropriate setting of the life-time parameter is difficult and requires careful calibration of system behaviour [17]. Our approach eliminates these problems without sacrificing coverage. We propose, and study, a family of protocols which preserve the topology-independent nature of flooding, while being able to achieve coverage levels arbitrarily close to 1, for any node density. Of course a specific high coverage cannot be guaranteed in any given instance, but can be expected with high probability. These protocols are based on a notion of ‘encounter’, and are controlled by an ‘encounter threshold’ parameter. The cost paid for a high coverage is an increase in the message traffic, since messages are broadcast more than once by each node. Under certain simplifying assumptions, it is shown that to achieve a coverage close to 1 in a network with n nodes, the total average number of broadcasts per message is on the order of O(n ln n). This is a moderate increase on the O(n) broadcasts carried out in flooding. The propagation time of a message is on the order of O(ln n). Thus, these protocols are particularly recommended for low density, highly mobile networks where reasonable delays can be tolerated.

At higher densities, our protocols, in their simple form, exacerbate the broadcast storm problem. To remedy this, we use existing and new optimisations which are efficient in reducing the number of broadcasts, without reducing coverage significantly and without requiring topology information.

Various aspects of the protocols’ performance are examined by simulation. We have also conducted an experiment (similar to the one reported in Hui et al. [15]), where performance measurements were taken from a real-world ad-hoc network.

The model, and the message propagation protocols, are described in Section 2. Some analytical results concerning the propagation time and the number of broadcasts are obtained in Section 3. The problem of improving the efficiency of the protocol by reducing the number of redundant broadcasts is addressed in Section 4. In Section 5, the protocol is generalised in order to handle the propagation of multiple messages in parallel. The outcomes of a number of simulation experiments with the basic, improved and generalised protocols, and with two different mobility patterns, are presented in Section 6. The validation exercise involving a real-life ad-hoc network is described in Section 7. Section 8 summarises the results obtained and outlines avenues of further enquiry.

Section snippets

The model

The system under consideration consists of n mobile nodes which move within a given terrain. The nodes communicate with each other using wireless technology, but without any fixed network infrastructure support. That is, the nodes themselves are the sources as well as the forwarders of the message traffic, and thus form a mobile ad-hoc network. Each node has a unique identifier (MAC or IP address). It is assumed that nodes do not run out of power and do not fail; however, due to their mobility,

Analytical approximation

In this section, we concentrate on evaluating the ability of EG(τ) to achieve high coverage, In order to make the model tractable, we assume the following:

  • The overheads of collision resolution are negligible.

  • Hello signals are sent and monitored at the MAC level; the information necessary to maintain the neighbourhood list is obtained at no extra cost to the higher level protocol.

  • Encounters last long enough for a message to be received, i.e. the processing and propagation times of hello and

Protocol improvements

Encounter propagation involves a trade-off between coverage and propagation overheads. The larger the value of τ, the higher the coverage achieved, but also the higher the number of redundant broadcasts (a broadcast is redundant if it does not enlarge the set of nodes that have already received the message). Each broadcast consumes power, shortens the battery operative period and, by increasing channel traffic, increases the likelihood of collisions. It is therefore important to keep the number

Multiple messages

In this section we address some of the issues that arise when more than one propagation, of messages originating at different nodes at different times, may overlap. To achieve such multiple propagations, each node must maintain a buffer of all messages that it has received, together with the corresponding counts indicating how many times each message has been broadcast. A simple generalisation of the EG(τ) protocol, where each node can keep track of up to M messages in the process of

Simulation results

A number of simulation experiments were carried out, aimed at evaluating the effect of various parameters on the performance of the protocols described in the previous sections. The following factors were kept fixed:

The terrain is a square of dimensions (1000 m) × (1000 m). The number of nodes is kept fixed at n = 64. The node density (defined as the average number of nodes within a circle of radius equal to the wireless range) is varied by altering the wireless range. Values for the density used

Experiments with a real-life MANET

Some of the assumptions we have made in this paper, and incorporated in the simulations described so far, may be difficult to defend in the context of a real-life ad-hoc network. Among the issues that have not been addressed here but may affect performance are:

Fading and transient network links [4]: Once a connection has been established between two nodes, even without mobility, the ability to transmit between them is not constant. Successful transmission of a packet over a wireless network

Conclusions

The main contributions of this paper can be summarised as follows:

  • 1.

    Introduction of the family of encounter propagation protocols (Section 2).

  • 2.

    Mobility-independent estimate for the value of τ that achieves high coverage (Section 3, Eq. (6)).

  • 3.

    Improvements to the protocol aimed at reducing the number of redundant transmissions (Section 4). In particular, a combination of RAD and α-reduction is both simple and efficient, and hence is worth implementing.

  • 4.

    An efficient generalisation to multiple

David Edward Cooper received his B.Sc. (Hons) Computing Science in 2001 and his M.Sc. System Design and Internet Applications in 2002 both at Newcastle University, UK. Continuing at Newcastle he received his Ph.D. in December 2008 and is currently a Research Associate of the University. His current research interests include Mobile Ad Hoc Networks and Dynamic Deployment of Web Services.

References (28)

  • A. Khelil et al.

    Hypergossiping: a generalized broadcast strategy for mobile ad hoc networks

    Ad Hoc Networks

    (2007)
  • M. Balazinska, P. Castro, Characterizing mobility and network usage in a corporate wireless local-area network, in:...
  • A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, J. Scott, Pocket switched networks: Real-world mobility and its...
  • C. Chiang, M. Gerla, On-demand multicast in mobile wireless networks, in: Proceedings of IEEE ICNP, 1998, pp....
  • K.W. Chin et al.

    Implementation experience with MANET routing protocols

    ACM SIGCOMM Computer Communication Review

    (2002)
  • D.E. Cooper, P. Ezhilchelvan, I. Mitrani, A real world evaluation of encounter gossip, Technical Report Series,...
  • D.E. Cooper, P. Ezhilchelvan, I. Mitrani, High coverage broadcasting for mobile ad-hoc networks, in: The Third IFIPTC6...
  • D.E. Cooper, P. Ezhilchelvan, I. Mitrani, E. Vollset, Optimization of encounter gossip broadcasting in mobile ad-hoc...
  • D.E. Cooper, P. Ezhilchelvan, I. Mitrani, E. Vollset, Optimization of encounter gossip propagation in mobile ad-hoc...
  • JJ Garcia-Luna-Aceves, E.L. Madruga, A multicast routing protocol for ad-hoc networks, in: Proceedings of IEEE INFOCOM,...
  • T. Gopalsamy, M. Singhal, D. Panda, P. Sadayappan, A reliable multicast algorithm for mobile ad hoc networks, in:...
  • M. Grossglauser et al.

    Mobility increases the capacity of ad hoc wireless networks

    IEEE/ACM Transactions on Networking (TON)

    (2002)
  • J. Hahner, C. Becker, K. Rothermel, A protocol for data dissemination in frequently partitioned mobile ad hoc networks,...
  • T. Henderson, D. Kotz, I. Abyzov, The changing usage of a mature campus-wide wireless network. in: Proceedings of the...
  • Cited by (7)

    • Rapid selection and dissemination of urgent messages over delay-tolerant networks (DTNs)

      2020, Advances in Delay-Tolerant Networks (DTNs): Architecture and Enhanced Performance
    • Rapid selection and dissemination of urgent messages over delay-tolerant networks (DTNs)

      2014, Advances in Delay-tolerant Networks (DTNs): Architecture and Enhanced Performance
    • Fair comparison of gossip algorithms over large-scale random topologies

      2012, Proceedings of the IEEE Symposium on Reliable Distributed Systems
    • Decentralised communication in autonomous agent swarms

      2012, Proceedings - 26th IEEE International Conference on Advanced Information Networking and Applications Workshops, WAINA 2012
    • Performance evaluation of epidemic broadcast with directional antennas in vehicular ad-hoc networks

      2011, Proceedings - 11th IEEE/IPSJ International Symposium on Applications and the Internet, SAINT 2011
    View all citing articles on Scopus

    David Edward Cooper received his B.Sc. (Hons) Computing Science in 2001 and his M.Sc. System Design and Internet Applications in 2002 both at Newcastle University, UK. Continuing at Newcastle he received his Ph.D. in December 2008 and is currently a Research Associate of the University. His current research interests include Mobile Ad Hoc Networks and Dynamic Deployment of Web Services.

    Paul Devadoss Ezhilchelvan holds a Personal Readership in Distributed Computing in the School of Computing Science, Newcastle University, from where he also received his Ph.D. in Computing Science in 1989. His research interests are primarily in the areas of dependability and fault-tolerance, distributed algorithms, group communication protocols, fair-exchange protocols, distributed agreement, simulations and performance evaluation. His research investigations cover both wired networking systems and mobile wireless systems. He was the principal architect of VOLTAN fault-tolerant systems and NewTOP group communication systems.

    Isi Mitrani studied Mathematics at the Universities of Sofia and Moscow, and Operations Research at the Technion, Haifa. He has spent most of his academic life at the University of Newcastle, where he is now an Emeritus Professor. His research interests are in the areas of probabilistic modelling, performance evaluation and optimisation. Publications include 7 books, more than 100 journal and conference papers and one patent.

    View full text