•   When: Monday, November 30, 2020 from 10:00 AM to 12:00 PM
•   Speakers: Katherine (Raven) Russell
•   Location: Virtual
•   Export to iCal

Abstract

Information Centric Disruption Tolerant Networks (ICDTNs) have been forwarded as a new powerful approach to provide effective data and information sharing when existing communication infrastructures are degraded or destroyed. ICDTNSs can assist with communication infrastructure collapse and disaster response and aid is spreading vital information in a fast and reliable manner. This dissertation investigates the construction of very large (city-scale) ICDTNs without infrastructure support for such scenarios.

I first present a new, scalable architecture for ICDTN construction using a hybrid approach with Named Data Networking (NDN) and Delay/Disruption Tolerant Networking (DTN) which avoids the need to heavily modify existing NDN and DTN implementations. I analyze different design choices for routing and location awareness and show that a geographically aware architecture is particularly effective for handling mobility at scale in an ICDTN. This conclusion is based upon evaluating ``geographical'' and ``node'' aware NDN implementations under a number of mobility models, network sizes, DTN forwarding methods, and caching in disaster scenarios. The results demonstrate the scalability and effectiveness of this architecture with the appropriate selection of geographical information, name advertisement, and caching.

The effectiveness of an ICDTN in an infrastructureless environment is largely dependent upon efficient and scalable data advertising. Data advertising is the process by which hosts are informed about available data in the system and how to request this data. I address this issue by exploring the use of random linear network coding.  My approach describes how to encode, decode, and use NDN advertising structures in an ICDTN environment, and I compare my solution to standard flooding techniques under a variety of network sizes, communication conditions, node densities, and mobility patterns. The results show that random linear network coding is highly scalable and can significantly decrease the time for advertisement message delivery.

I then develop a general-purpose probabilistic model for the data advertisement problem. Using this model, I analyze three protocols: a naive flooding mechanism, a network coding solution, and an improved flooding protocol inspired by the network coding approach. By measuring several key performance metrics and comparing the analysis results with empirical data for a variety of network topologies in disrupted environments, I show that network coding, and network coding inspired techniques, can be used to greatly reduce the minimum receive times and shorten the announcement period. My probabilistic model provides framework for analyzing future advertising protocols. 

Finally, using Agent-Based Modeling (ABM) I evaluate the performance of my hybrid architecture and data advertisement protocols in a metro-level disaster response scenario. With over a million devices modeled, these results expose challenges not currently addressed in the literature that are unique to large-scale ICDTNs, including the interplay of NDN parameters, the effects of local caching, and the challenges with mass human movement. Experimentally, the hybrid architecture with the geographical-aware NDN setup can easily scale to deliver up-to-date information in an ad hoc manner in such a disaster scenario.

My work paves a way forward for future real-world implementations of city-scale infrastructureless ICDTNs by providing a ground-up approach to the problem. It addresses the initialization of the network during the data advertisement process, the architecture of the network needed to handle information distribution in disrupted environment, and an ABM approach to studying these networks at scale.