Content Distribution over Named-Data Networks
Chavoosh Ghasemi · UA Campus Repository (The University of Arizona) · 2020
Radical changes in Internet usage during the past few decades have created a big semantic gap between what the network provides (i.e., transport service) and what services and applications want (i.e., contents). The current Internet is an enormous transport network connecting two endpoints to each other and moves bits from one endpoint to another with no sense of what content has been requested. This network, although provides applications with transport services, can barely support content distribution at scale. The urgent need for a large-scale content distribution solution has resulted in the emergence of content networks. A content network understands what content has been solicited and tries to retrieve it from anywhere in the network, whether a cache or the origin (a.k.a. content producer). Such a network is a natural fit for today’s Internet as modern applications/services mostly ask for certain contents instead of accessing particular machines. To realize content networks within TCP/IP architecture with minimal changes to the Internet content distribution networks (CDNs) have been deployed. However, as the scale increases these networks are also facing challenges such as system complexity, resource footprint, and content security. The emerging Information-Centric Networking (ICN) technology, Named Data Networking (NDN) in particular, offers a network architecture that directly supports content distribution without requiring an overlay to bridge the gap between network services and application needs. This enables ICN/NDN to be potentially capable of supporting large-scale content distribution with less system complexity and resource footprint than TCP/IP-based solutions. In this dissertation, for the first time to the best of our knowledge, we scrutinize the content network built by NDN technology at the large-scale and compare it to TCP/IP-based solutions in the wild. We conduct real-world experiments to compare the standard deployment of NDN (i.e., the global NDN testbed) and two well-known CDNs (Akamai and Fastly) by deploying an NDN-based adaptive video streaming service, called iViSA. We evaluate these networks in terms of caching and retrieving static contents through streaming hundreds of videos from four different continents for two weeks. After showing where the current deployment of NDN stands today and speculating its main practical challenges, we propose iCDN, a scalable, resilient, and high-performance content network using NDN technology. Also, to help deploy native data-centric communication at the global scale we present NameTrie, a highly efficient data structure for handling millions of names on common hardware commodities.