IEEE Access (Jan 2024)

Modeling and Evaluating a Cache System in ICN Routers Using a Programmable Switch and Computers

  • Junji Takemasa,
  • Yuki Koizumi,
  • Toru Hasegawa

DOI
https://doi.org/10.1109/ACCESS.2024.3484588
Journal volume & issue
Vol. 12
pp. 154426 – 154442

Abstract

Read online

Information-centric networking (ICN) is one of promising networking architectures to replace IP because its notable feature, in-network caching, is expected to reduce about a one-third of the forever increasing Internet traffic. However, the existing ICN router implementations with caches cannot come with the rapid increase of terabit-scale network bandwidth. An implementation on hardware-based router platforms like programmable switches is promising, but due to their limited memory capacity, cache stores should be located at external devices like computers. This makes packets being sent from and to the switch to cache them at the computers, which increases the number of ports for connecting them called external ports. That is, the number of ports for connecting the switch and external networks is reduced, and as a result, the packet forwarding rate is degraded. We analytically and experimentally show that a naive implementation decreases both the packet forwarding rate and the number external ports to almost half in the condition that the cache hit ratio is about 30%. To solve this bottleneck, this paper proposes the two algorithms, the cache admission and lazy response ones, to reduce the number of packets sent between the switch and the computers. In order to validate them, we evaluate the forwarding rate improvement by developing an analytical model and a prototype ICN router implementation on a programmable switch with computers. The evaluation shows that the proposed algorithms almost double the forwarding rate compared to a naive implementation without the proposed algorithms under realistic request patterns to data packets. The paper’s contributions are two-fold: realistic implementation and comprehensive performance evaluation. First, we implement a terabit-class ICN router with cache functionality on a Tofino switch which provides 3.2 terabit/s (Tbps) forwarding rate with two pipelines. The implemented router achieves 1.075 Tbps forwarding rate in the condition that just one pipeline of the Tofino switch is used and that the cache hit ratio is about 30%. As far as we know, this prototype implantation is one of the first full-fledged ICN routers on a Tofino switch. Second, the performance of the proposed router is both analytically and experimentally evaluated, and the both results show the similar packet forwarding rate under realistic request patterns following the Zipf distribution, which the Internet traffic is believed to follow. Finally, the paper is extended from the conference paper so that the proposed algorithms are evaluated both analytically and experimentally, whereas they are mainly evaluated by simulation in that paper.

Keywords