Content Dissemination Protocols in Hybrid Wireless Networks

Wireless hybrid networks have become quite popular with the evolution of several devices with multiple networking interfaces. Each of these interfaces (and the associated networks) have significantly different characteristics, hence it is important to understand how best to utilize and operate in the presence of multiple networks. Smartphones are a popular example of devices with multiple networking interfaces, e.g. WiFi and cellular. As these devices become popular with soldiers in battlefields for communications, an important question is to address the issue of how best to utilize the multiple networks while minimizing delay in content delivery and maximizing the utility of these networks. In this paper, we explore the tradeoffs between the use of cellular and WiFi interfaces in the context of a network of mobile nodes (patrolling a specified area). We consider a network model in which nodes use WiFi in ad-hoc mode (forming a MANET) and exchange data when they come in contact with each other, whereas the cellular link provides direct connectivity. We address the problem of disseminating content to a group of mobile nodes (possibly the entire network) in the presence of a cellular and ad-hoc WiFi networks. Cellular links are bandwidth-constrained and expensive, but provide low delay for data dissemination. On the other hand, ad-hoc WiFi networks offer higher bandwidth and are inexpensive, but could incur high data delivery delays due to intermittent node connectivity. We develop a protocol for achieving this low delay and low cost data delivery that employs machine learning techniques in a novel manner and evaluate the efficacy of this protocol on a dataset with 100 mobile nodes (patrolling a specified area). We show through extensive simulations based on real-world mobility patterns of the 100 nodes that the bandwidth consumed on the cellular link can be reduced by about 50% on an average while not increasing the delay incurred (in data delivery) significantly.