Elsevier

Ad Hoc Networks

Volume 45, 15 July 2016, Pages 34-46
Ad Hoc Networks

Opportunistic content diffusion in mobile ad hoc networks

https://doi.org/10.1016/j.adhoc.2016.02.022Get rights and content

Abstract

Opportunistic wireless content sharing via Mobile Ad hoc NETworks (MANETs) can increase throughput, lower latency, extend network coverage and reduce load on infrastructure. While the benefits of content diffusion clearly depend on the underlying movement dynamics and content demand, the impact of these factors on diffusion remains largely unexplored. We analyze content sharing potential based on device encounters inferred from a large multi-site wireless LAN trace. We explore the impact of time, location, and number of sources on diffusion, finding that contexts with higher activity generally promote faster diffusion, while additional content sources improve diffusion mainly in the short-term. We then apply real-world demand patterns from a popular campus maps application to content diffusion simulations. We find that up to 70% of map requests could theoretically be served from the peer network over the first 12 h. Finally, our analysis of the impact of trace uncertainties and individual device variation on diffusion potential reveals large differences based on the selected assumption and chosen source devices. We discuss these results and their implications for content-diffusion in MANETs.

Introduction

Enabling wireless user devices to directly share common-interest content is a conceptually attractive approach to enhancing wireless networks. Each user device caches content retrieved from the infrastructure and makes it transparently available to colocated peers, either pre-emptively or on demand. Devices’ content demands are preferentially served from a nearby peer with the infrastructure serving as a fallback when a cached copy is unavailable. The potential benefits of such a scheme include higher throughput, lower latency, greater spectrum reuse, extended network coverage and reduced load on infrastructure.

We present a mobile map sharing application as a motivating example. Suppose User A is using their mobile device to navigate a geographic region after having downloaded the region’s map from the infrastructure (e.g. a cell tower or wireless access point). Now suppose User B enters the same region and encounters User A. User A proceeds to pre-emptively share the map data with User B. Shortly afterwards, User B would also like to view a map of the region. Rather than having to retrieve the mapping data from the infrastructure, User B already has a local copy available received earlier from User A. We highlight several potential benefits of this peer sharing:

  • Being in close geographic proximity allows the devices to transmit at lower power, reducing battery consumption and increasing opportunities for spectrum reuse in adjacent areas.

  • User A and B can establish a short-range dedicated connection, increasing throughput. This is particularly important if User B were to retrieve the map on demand, rather than receiving it pre-emptively.

  • The devices can communicate with very low latency as a result of the short-range nature of the connection and because the devices are not contending with other devices for access to the infrastructure. Again, this is important for on-demand retrieval.

  • If User B is not in range of the infrastructure, User A effectively extends User B’s coverage by making otherwise unreachable content available.

  • Finally and in many cases most importantly, load has been taken off the fixed wireless infrastructure. Wireless infrastructure and cellular data infrastructure in particular is often viewed as being in a perpetual state of underprovision. Partially offloading content delivery from the infrastructure onto a Mobile Ad hoc NETwork (MANET) may prove a useful strategy for reducing the necessary cost or frequency of infrastructure upgrades.

Continuing the maps example, assume that some time later User A transitions to a new geographic region. As a result of A’s mobility, maps of the prior region are now available to devices in the new region. This is an example of how content may spread with the aid of device mobility.

We have presented mapping as just one motivating example of MANET-based content sharing via diffusion. The use cases of content diffusion however generalize to any application premised on or enhanced by the ability to move content quickly and efficiently. Content diffusion may prove particularly useful for other applications which like maps exhibit locality of reference [1] in content interests, i.e. content interests tend to be spatially and/or temporally correlated. This includes web content, app content and even personal area networks (PANs) where a single user carries multiple cloud-connected devices synchronizing identical data.

Though wireless peer-to-peer (P2P) content sharing is an intellectually attractive approach to improving network efficiency and performance, a lacuna exists in the literature around real-world parameters influencing content diffusion potential. Existing works [2], [3] explore some facets of epidemic content diffusion including the resulting network topologies and diffusion potential under various constraints on participation. Our earlier work in [4] provides a preliminary examination of how site, time of day, day of week, number of content sources and empirical patterns of content demand influence content diffusion potential in wireless LANs. In the present paper we build on our prior work by analyzing the impact of uncertainty and variation in trace-driven diffusion simulations. We find diffusion potential to be relatively sensitive to the assumptions chosen to compensate for inherent timing uncertainties in wireless LAN traces. We also find a relatively large amount of variability in diffusion potential between individual content source devices. We discuss currently accepted assumptions of the research community as they pertain to inferring device encounters and highlight why verifying the validity and then perhaps improving these assumptions would be beneficial.

The following section covers related work. Section 3 provides background information on the area of content diffusion and formally defines how device encounters are inferred from wireless LAN traces. Our primary wireless LAN trace from a large university campus is described in Section 4, along with its associated uncertainties in session timestamps. Our first set of simulations analyze universal diffusion on the empirical trace, i.e. how quickly an arbitrary piece of content might spread throughout a network. These simulations are described in Section 5 and the results are presented in Section 6. We then focus on a realistic application-specific use-case for content diffusion in Section 7—diffusing electronic maps based on the LAN trace and on empirical usage statistics from a university navigation app. Section 8 provides a discussion of our findings regarding the impact of trace uncertainties and presents avenues for future work. Section 9 concludes the paper.

Section snippets

Related work

Our work fits broadly into the existing body of research around MANET [5]communications and Delay Tolerant Networking (DTN) [6]. Though present-day device and protocol support for seamless device-to-device communication is somewhat deficient, we are particularly motivated in our analysis by promising next generation protocols like Content-Centric Networking (CCN) [7]. The pertinent feature of CCN (and similar protocols) is enabling trustworthy content to be retrieved from untrusted hosts.

Most

Opportunistic mobile content diffusion

Opportunistic mobile content diffusion refers to the dissemination of content directly between mobile devices during incidental encounters, i.e. where and when opportunities naturally arise. Content may originate directly from a device or have been downloaded from an infrastructure network at an earlier point in time. For example, a sensor reading may originate from a mobile device, while a cached web page originates from an Internet-connected infrastructure network. Once one or more mobile

The UQ trace

The UQ trace is a record of all IEEE 802.11 (Wi-Fi) Access Point (AP) sessions collected from the multi-site University of Queensland (UQ) wireless network between Nov. 27–Dec. 11, 2012. The trace contains 549,002 sessions from 23,931 unique MAC addresses connecting to 3081 APs across 24 discrete geographic sites. Sites include university campuses, hospitals, research stations and AP installations at other UQ-affiliated locations throughout the state of Queensland, Australia. Each record in the

Simulation overview

Using our empirical traces, we perform multi-site, multi-source simulations for a variable number of source devices, variable diffusion start times and under both pessimistic and optimistic session length assumptions. Our simulation models ideal content diffusion by means of Discrete Event Simulation (DES) implemented as a set of custom Shell, Python and Go scripts. In total we perform 10,000 universal content diffusion simulations. This entails simulating all combinations of 5 sites, 5

Results presentation overview

Throughout this section, we refer to Figs. 3–6 to illustrate our findings.

Fig. 3 is a heatmap of the time taken for the unreachable ratio to drop to 50% under all combinations of the simulated parameters. The purpose of Fig. 3 is to provide a coarse summary measure of diffusion performance—the time taken for diffused content to reach half of all devices in the simulated network.

Figs. 4 and 5 depict the unreachable ratio over time for each site using different combinations of diffusion start

Simulating application-specific diffusion

In this section we examine a concrete use case of information diffusion—sharing electronic maps. Our simulations draw upon both the UQ trace and the JCUNav trace (described next) to model diffusion of maps between wireless devices. From the UQ trace we use the same set of sessions and inferred encounters used earlier in our universal diffusion simulations. We then project the daily and hourly usage patterns from the JCUNav trace (Figs. 7 and 8) onto the UQ trace to simulate demand for maps

Discussion and future work

The results presented in this paper elucidate a number of tangible factors influencing rates of information diffusion. However, our comparison of diffusion potential under optimistic and pessimistic assumptions also highlights diffusion’s sensitivity to trace uncertainties. Some traces like the UQ trace embed uncertainties regarding session start and end times which are the result of periodic rather than instantaneous sampling of connected devices. Other forms of uncertainty however are more

Conclusion

Our analysis of MANET-based content diffusion reveals several important factors influencing diffusion potential. Firstly, the rate at which content spreads throughout a network is highly site-dependent, even across sites of the same type (university campuses) and even when the trace collection is controlled for both network type and collection period. Secondly, the time at which content is introduced into the MANET greatly influences the success of information diffusion over the short-term. In

Acknowledgments

This work is supported in part by an Australian Government Australian Postgraduate Awards scholarship and Commonwealth Scientific and Industrial Research Organization Office of the Chief Executive scholarship.

Bryce Thomas received the bachelor’s degree in Information Technology from Central Queensland University, Australia and honors in Information Technology from James Cook University, Australia. Bryce is currently an Information Technology PhD candidate at James Cook University.

References (22)

  • N. Aschenbruck et al.

    Trace-based mobility modeling for multi-hop wireless networks

    Comput. Commun.

    (2011)
  • P. Holme et al.

    Temporal networks

    Phys. Rep.

    (2012)
  • P.J. Denning

    The locality principle

    Commun. ACM

    (2005)
  • W. Hsu et al.

    On nodal encounter patterns in wireless LAN traces

    IEEE Trans. Mob. Comput.

    (2010)
  • J. Su et al.

    User mobility for opportunistic ad-hoc networking

    Proceedings of the Sixth IEEE Workshop on Mobile Computing Systems and Applications

    (2004)
  • B. Thomas et al.

    Content diffusion in wireless MANETs: The impact of mobility and demand

    Wireless Communications and Mobile Computing Conference (IWCMC), 2014 International

    (2014)
  • D. Johnsort

    Routing in ad hoc networks of mobile hosts

    Mobile Computing Systems and Applications, 1994. WMCSA 1994. First Workshop on

    (1994)
  • K. Fall

    A delay-tolerant network architecture for challenged internets

    Proceedings of the 2003 conference on Applications, technologies, architectures, and protocols for computer communications

    (2003)
  • V. Jacobson et al.

    Networking named content

    Proceedings of the 5th international conference on Emerging networking experiments and technologies

    (2009)
  • N. Eagle et al.

    Reality mining: sensing complex social systems

    Pers. Ubiquitous Comput.

    (2006)
  • P. Hui et al.

    Pocket switched networks and human mobility in conference environments

    Proceedings of the 2005 ACM SIGCOMM workshop on Delay-tolerant networking

    (2005)
  • Bryce Thomas received the bachelor’s degree in Information Technology from Central Queensland University, Australia and honors in Information Technology from James Cook University, Australia. Bryce is currently an Information Technology PhD candidate at James Cook University.

    Raja Jurdak is a Principal Research Scientist at CSIRO, where he leads the Distributed Sensing Systems Group. He has a PhD in Information and Computer Science at University of California, Irvine in 2005, an MS in Computer Networks and Distributed Computing from the Electrical and Computer Engineering Department at UCI (2001), and a BE in Computer and Communications Engineering from the American University of Beirut (2000). His current research interests focus on energy-efficiency and mobility in networks. He has over 100 peer-reviewed journal and conference publications, as well as a book published by Springer in 2007 titled Wireless Ad Hoc and Sensor Networks: A Cross-Layer Design Perspective.

    Ian Atkinson is a Tropical Leader and Director of the eResearch Centre at James Cook University. As Director of the JCU eResearch Centre he has the privilege of working with researchers to apply new and constantly changing ICT tools and methods to multiple research domains. He has a long-standing interest in eResearch methods, tools, scientific data management and user interfaces for HPC tools. He is also actively involved in researching how new systems and software that connect the physical and virtual worlds, particularly focusing on environmental monitoring with sensor networks.

    View full text