Preprint / Version 1

An HTTP/3 CDN-Based Object-Centric Architecture for Real-Time Media Delivery

##article.authors##

DOI:

https://doi.org/10.31224/8175

Keywords:

CDN, HTTP/3, QUIC, Real-Time Media Delivery, Object-Centric, Low-Latency Delivery

Abstract

In recent years, low-latency real-time media delivery technologies represented by WebRTC and WebTransport have become widespread. However, many of these technologies assume a session-oriented communication model, and face challenges in resilience to frequent disconnections and reconnections in mobile environments, as well as cost efficiency during large-scale delivery. On the other hand, while CDNs provide high scalability and stable transfer performance, they have traditionally been considered unsuitable for real-time applications. This study redefines real-time media delivery not as session-oriented communication, but as a delivery problem of continuously updated objects. In the proposed method, the latest media data is delivered as objects using long-lived connections over HTTP/3, and state management is decoupled from the transport layer. As a result, even when disconnections occur, delivery can be immediately resumed through offset-based retrieval without requiring reconstruction of session state. Furthermore, by assuming delivery through CDNs, the proposed method achieves a linear cost model proportional to transfer volume even as the number of end users increases. In evaluation experiments, Time To First Audio Byte (TTFAB) and operational costs were measured and compared with conventional WebRTC and SFU configurations. The results showed that the proposed method achieved approximately 3x faster TTFAB (0-RTT, 180ms) compared to WebRTC P2P (best case, 500 to 600ms), with an order of magnitude difference when candidate fallback occurs in NAT environments. Advantages in cost efficiency during large-scale delivery were also confirmed. This study presents a practical architecture for real-time media delivery through object-centric design using HTTP/3 CDN.

Downloads

Download data is not yet available.

Downloads

Posted

2026-09-10