On Content Indexing for Off-Path Caching in Information-Centric Networks
Suzan Bayhan, Liang Wang, Jörg Ott, Jussi Kangasharju, Arjuna Sathiaseelan, Jon Crowcroft · 2016
A name resolution server (NRS) in an Information-Centric Network (ICN) can leverage the off-path copies in the network, which may not be accessible via content discovery mechanisms. Such capability is essential for an Autonomous System (AS) to avoid the costly inter-AS traffic for external content, to yield higher bandwidth efficiency for intra-AS traffic, and to decrease the data access latency for a pleasant user experience. However, these benefits come at the expense of storage and NRS update costs, for which scalability is paramount given huge number of contents. In this article, we argue that most of the benefits of an NRS can be realized by indexing only a small fraction of the requested content benefiting from the NRS the most. First, we model the cost of serving each content in the existence of an NRS and lack of it, considering content’s popularity, availability, size, and type. Then, we derive the optimal indexing decision under a given NRS size constraint by formulating an optimization problem that minimizes total cost for serving all requests within this AS. Our results suggest that an NRS tracking even only a tiny fraction of the most popular (external) content delivers most of the benefits of an NRS, e.g., lower inter-AS traffic, higher cache hit, and lower latency. While larger NRS provides slightly higher cache hits for small caches, the impact is more visible for larger cache capacity. In contrast to diminishing gains in cache hit, data latency decreases further with increasing NRS size owing to faster name resolution.