Date of Award
Winter 12-15-2015
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
Named Data Networking (NDN) is a recently proposed general-purpose network architecture that aims to address the limitations of the Internet Protocol (IP), while maintaining its strengths. NDN takes an information-centric approach, focusing on named data rather than computer addresses. In NDN, the content is identified by its name, and each NDN packet has a name that specifies the content it is fetching or delivering. Since there are no source and destination addresses in an NDN packet, it is forwarded based on a lookup of its name in the forwarding plane, which consists of the Forwarding Information Base (FIB), Pending Interest Table (PIT), and Content Store (CS). In addition, as an in-network caching element, a scalable Repository (Repo) design is needed to provide large-scale long-term content storage in NDN networks.
Scalable NDN forwarding is a challenge. Compared to the well-understood approaches to IP forwarding, NDN forwarding performs lookups on packet names, which have variable and unbounded lengths, increasing the lookup complexity. The lookup tables are larger than in IP, requiring more memory space. Moreover, NDN forwarding has a read-write data plane, requiring per-packet updates at line rates. Designing and evaluating a scalable NDN forwarding node architecture is a major effort within the overall NDN research agenda.
The goal of this dissertation is to demonstrate that scalable NDN forwarding is feasible with the proposed data structures and algorithms. First, we propose a FIB lookup design based on the binary search of hash tables that provides a reliable longest name prefix lookup performance baseline for future NDN research. We have demonstrated 10 Gbps forwarding throughput with 256-byte packets and one billion synthetic forwarding rules, each containing up to seven name components. Second, we explore data structures and algorithms to optimize the FIB design based on the specific characteristics of real-world forwarding datasets. Third, we propose a fingerprint-only PIT design that reduces the memory requirements in the core routers. Lastly, we discuss the Content Store design issues and demonstrate that the NDN Repo implementation can leverage many of the existing databases and storage systems to improve performance.
Language
English (en)
Chair
Patrick Crowley
Committee Members
Roger Chamberlain, Roch Guerin, Raj Jain, Neil Richards, Jonathan Turner
Comments
https://doi.org/10.7936/K7JQ0Z92