Data-Centric Wireless Networks

Wireless networking is ubiquitous in IoT applications, edge computing, cellular and tactical networks, to name a few. However, communication over wireless media is inherently challenging because of packet collision and channel contention. Mobility exacerbates these challenges manifold by introducing frequent link breakage. Today's IP-based communication is built on top of the telephony model, assuming a stable end-to-end channel between a specific sender-receiver pair identified by addresses. As a result, traditional IP packet forwarding in the network and transport control exhibit high packet loss and latency. Thus they employ complex protocols for reliability, albeit still struggle to offer high throughput or success rate. Named Data Networking (NDN) is an emerging Internet architecture based on Information-Centric Networking (ICN) philosophy. The network stack is built from the ground up, focusing on the application's needs. NDN uses names to identify and find a data packet or information in the network rather than finding a specific address. This slight paradigm shift unlocks many advantages without intricate protocol management, such as in-network caching for better reliability, built-in multicast support without complex tree maintenance, loop detection, out-of-order data retrieval, and per-packet security rather than securing a single channel. Such advantages promise to support wireless networks better. However, existing works lack the comprehensive analysis and pitfalls the architecture may exhibit under lossy and unstable communication conditions. In this work, I focus on NDN's potential advantages and challenges by analyzing the architecture from the application to the link-adaptation layer and propose suitable protocol designs. An application-level Dynamic Interest Lifetime (DIL) technique reduces data redundancy on application retransmission. At the same time, using a Congestion Window Limit (CWL) caps the upper bound of congestion at the transport. Together, they help reduce channel contention and packet collision, improving application throughput. Next, a Data-centric Ad-hoc Forwarding (DAF) at the network layer does away with a separate routing control plane and learns a sub-optimal shorter path toward a data node. Doing so significantly reduces learning latency compared to similar IP routing in ad-hoc networks. With NDN's multicast and caching, DAF offers a better success rate than AODV and shows NDN is a better fit than IP in mobile ad-hoc networks (MANETs). I also propose a hybrid probabilistic forwarding (HyPr) to support delay-tolerant networking through NDN's built-in caching capabilities. HyPr improves in-network resource usage like limited cache capacity compared to an epidemic routing by forwarding Interest packets through probability calculation based on neighborhood contact count. It lowers redundant data, cache eviction, and total network transmission events. Finally, an Interest bundling technique named BLEnD at the link adaptation layer improves channel usage for data packets by retrieving multiple data with a single Interest packet. Doing so enhances application throughput over a single wireless hop while hiding the Interest bundling and unbundling from the upper layers of the NDN stack and avoids any potential breakdown of the stateful forwarding