Optimization and modeling of content delivery algorithms for dense mobile social networks

Abstract

Due to the increase in mobile phones, Ad-hoc Mobile Social Networks (MSNs) are gaining serious research and industry attention, i.e. information exchange in an opportunistic fashion: whenever two people meet, the one whose content is most recent pushes (using Bluetooth or Adhoc wireless) to the one whose content is outdated.

Mobile users can deliver content via existing cellular networks. However, there are several benefits of using such Ad-hoc MSNs. These exchanges can be used to extend the network’s coverage and improve the service provided to any user whose access to the wireless network is intermittent. Most importantly, such ex- changes can increase network’s capacity. In particular, by utilizing the bandwidth of connections between users, the service providers can support more subscribers at a lower cost. Moreover, MSNs can help in content filtering as usually people only want to exchange profile, exchange data, get opinions, trust information, take recommendations, event notification, and news etc from the people they trust.

The contributions of this thesis are two folds. First, we propose time-critical con- tent delivery algorithms for ad-hoc MSNs. The current content delivery algorithms for MSNs do not support time-critical delivery of content. As in these approaches an important behavior of MSNs is ignored i.e. predictability of the time of upcom- ing individual encounters/re-encounters. Considering predictable encounter-time in the protocols gives time assurance of message delivery for time critical application (such as: emergency situations notification, traffic congestion information, short event notice etc). Therefore in this thesis we address the research question, How to exploit the predictable behavior of MSNs to disseminate time critical dynamic content?

Secondly, while testing our content delivery algorithms we identified two major shortcomings of the current simulation models that either that are based on random models or use mathematical distribution (i.e. power-law distribution) to model encounters-time and contact-durations. We argue that these simulations models are not suitable for simulating MSNs because, Firstly the social encounters are not random and are based on some social purpose or social role and Secondly, modelling these encounters by normalizing using mathematical distributions loose important details about time of the day of an encounter and upcoming contact durations. Therefore, in this thesis we prove that by involving real people in life-simulation games can provide us with mobility patterns that are close of the real-life like encounter patterns and are more realistic to validate content delivery algorithms for MSNs. The performance and costs of our proposed content delivery algorithms was evaluated on our proposed simulation environment through empirical and numer- ical analysis. The analysis shows the feasibility of the algorithms for time-critical content delivery in MSNs.