Concordia Research: Co-designing Network, Edge, and AI to Cut Latency
An abstract for an upcoming research seminar details a new cross-layer architecture designed for next-generation interactive systems. The research, led by Dr. Abdelhak Bentaleb of Concordia University, argues that protocols like QUIC are insufficient and proposes a unified stack where networking, edge, and AI compute are co-designed. The approach uses AI for QoE-aware congestion control, multi-CDN orchestration, volumetric video streaming optimization, and pipelining video transport into Multimodal LLM processing to reduce latency.
Key Takeaways
- The proposed "cross-layer" architecture uses AI for QoE-aware congestion control and multi-agent reinforcement learning for real-time Multi-CDN orchestration.
- For immersive media, the work includes a full volumetric (6DoF) video ecosystem featuring a browser-native player and a large-scale dataset of user navigation traces.
- A new method pipelines progressive video transport directly into incremental Multimodal LLM processing, designed to eliminate GPU cold-start overhead.
- The research is supported by industry partners including InterDigital, Ericsson, and TELUS.
Why It Matters
This research argues that incremental improvements to protocols like QUIC are insufficient for delivering ultra-low latency. The core problem, termed the 'Blind Loop,' is framed as a fundamental architectural issue requiring an integrated approach that connects the network directly to application QoE. While many vendors optimize within their respective silos—transport, CDN, or compute—this model advocates for a holistic system where all layers are co-designed. The involvement of partners like Ericsson and TELUS indicates carrier-level interest in these deep integrations. Watch for whether cross-layer telemetry concepts begin to influence transport protocol standards or experimental edge compute offerings.
Read full article at events.comp.nus.edu.sg