RXC: Runtime Extra Compression for Time-Critical Content Delivery
Abstract
Recent advances in user devices and computer vision techniques have significantly increased daily multimedia traffic. Much of this traffic is generated by delay-sensitive services, including video conferencing, cloud gaming, and augmented reality. Existing adaptive delivery methods are often client-driven and rely heavily on end-host feedback. Under time-varying bandwidth, this feedback can be delayed or noisy, often leading to conservative utilization or avoidable delay. This article introduces Runtime eXtra Compression (RXC), a framework that allows an in-path node to dynamically apply extra compression to multimedia content during delivery. RXC exploits local network and compute conditions that are not visible through end-host feedback, enabling the encoded content to be adapted during transmission as downstream conditions evolve. As one concrete realization, we present Dynamic Compression Control (DCC), a delay-aware control mechanism, together with a multi-level compression implementation based on a Dynamic Multi-level AutoEncoder (DMAE). To demonstrate the practical value of RXC, we consider a case study of cloud-offloaded multimedia delivery over an AI-RAN access network, where a RAN-side node decides whether to apply extra compression to cloud-originated content based on locally observed conditions before forwarding it over the downlink to the user. Our evaluation shows that RXC achieves a more favorable latency-quality tradeoff than sender-side compression baselines under dynamic network conditions.