Title: StarCDN: Moving Content Delivery Networks to Space
Author: William X. Zheng, Aryan Taneja, Maleeha Masood (University of Illinois & Urbana-Champaign), Anirudh Sabnis (Akamai Technologies), Ramesh K. Sitaraman (University of Massachusetts Amherst & Akamai Technologies), Deepak Vasisht (University of Illinois & Urbana-Champaign)
Scribe: Gao Han (Xiamen University)
Introduction
While constellations like Starlink provide global Internet access, existing terrestrial CDNs are not designed for their highly dynamic architecture. Satellites move rapidly, coverage areas change constantly, and ground-satellite links are costly bottlenecks. As a result, today’s CDN deployments offer poor performance for satellite users and fail to relieve congestion in the satellite backhaul. This issue is increasingly important as LEO constellations scale to thousands of satellites and millions of subscribers, creating a growing demand for efficient, satellite-native content delivery. To this end, this paper proposes StarCDN, a new CDN architecture that caches content in space to improve user experience and reduce ground-satellite bandwidth usage.
Key Idea and Contribution
The paper proposes a space-based content delivery network called StarCDN that is specifically architected to work with LSNs. It is to achieve effective in-orbit caching under the highly dynamic characteristics of low-Earth orbit satellite constellations. The fundamental challenge in designing StarCDN lies in the orbital motion of satellites, which causes each satellite’s coverage area to change rapidly, serving vastly different regions within minutes. To address this, the paper introduces new consistent hashing and relayed fetching schemes tailored to LEO satellite networks.
First, it introduces a consistent hashing scheme that distributes cached objects across satellites in a way that minimizes redundancy and ensures availability, even as satellites move rapidly and their coverage areas shift. This prevents caches from becoming obsolete as satellites orbit the Earth.
Second, it employs a relayed fetching mechanism that allows content to be forwarded against the direction of orbital motion. By doing so, StarCDN keeps popular data close to regions of demand, counteracting orbital drift and maintaining cache locality for users.
Through these mechanisms, StarCDN makes space-based caching both feasible and efficient, improving performance for satellite users while reducing pressure on ground-satellite links.
Evaluation
The paper evaluates StarCDN with a trace-driven simulator that combines orbital motion models, cache replay, and real-world Akamai production traces. Using 5-day synthetic traces representing billions of requests, they compare StarCDN against naive LRU-based in-satellite caches and a static baseline. Results show that StarCDN achieves up to 80% reduction in ground-to-satellite bandwidth utilization and improves user-perceived latency by 2.5×. Additional experiments demonstrate robustness to satellite unavailability and strong cache hit rates across traffic classes (video, web, and downloads).
Question
Q1： Compared with traditional consistent hashing, StarCDN still needs duplicate content on multiple satellites. Do you know the quantitative redundancy?
A1: Yes, there is redundancy. With four buckets over about one thousand satellites, some duplication is inevitable. But this design keeps lookup within at most two hops, balancing efficiency and latency. If we reduce redundancy, more hops are needed, which leads to higher cache misses and latency. In our study, nine buckets already showed clear latency penalties.
Q2: Why did you use byte hit rate? Wouldn’t object hit rate be better?
A2: We also report object hit rate in the paper. But byte hit rate is more directly correlated with bandwidth savings, which is one of the key results we emphasize.
Personal Thoughts
StarCDN implements mechanisms for placing and fetching content in satellite-based caches to reduce user-perceived latency and optimize the utilization of ground-to-satellite links. By introducing a consistent hashing scheme adapted to orbital motion and a relayed fetching mechanism to maintain cache locality, the paper shows that caching in space is both feasible and beneficial. This offers a new approach to integrating satellite constellations with content distribution.