Towards Context-Aware Edge-Cloud Continuum Orchestration for Multi-user XR Services

The rapid growth of multi-user eXtended Reality (XR) applications, spanning fields such as entertainment, education, and telemedicine, demands seamless, immersive experiences for users interacting within shared, distributed environments. Delivering such latency-sensitive experiences involves considerable challenges in orchestrating network, computing, and service resources, where existing limitations highlight the need for a structured approach to analyse and optimise these complex systems. This challenge is amplified by the need for high-performance, low-latency connectivity, where 5G and 6G networks provide essential infrastructure to meet the requirements of XR services at scale. This article addresses these challenges by developing a model that parametrises multi-user XR services across four critical layers of the standard virtualisation architecture. We formalise this model mathematically, proposing a context-aware framework that defines key parameters at each level and integrates them into a comprehensive Edge-Cloud Continuum orchestration strategy. Our contributions include a detailed analysis of the current limitations and needs in existing Edge-Cloud Continuum orchestration approaches, the formulation of a layered mathematical model, and a validation framework that demonstrates the utility and feasibility of the proposed solution.

💡 Research Summary

The paper addresses the pressing need for efficient orchestration of multi‑user extended reality (XR) services over 5G/6G‑enabled edge‑cloud continuums. XR applications, ranging from entertainment and education to tele‑medicine, demand ultra‑low latency, high bandwidth, and real‑time synchronization among geographically dispersed participants. Existing orchestration mechanisms, while effective for single‑user or static workloads, fall short in handling the dynamic, context‑rich environment of multi‑user XR, where network conditions, device capabilities, user interaction intensity, and spatial consistency must be jointly considered.

To fill this gap, the authors formulate three research questions: (1) Are current edge‑cloud orchestration solutions sufficient for multi‑user XR? (2) Which parameters constitute the context needed for XR services, and how can they be categorized across architectural layers? (3) How can this context‑aware parametrisation improve orchestration decisions?

The paper’s contributions are threefold. First, it conducts a comprehensive literature and standards review (3GPP Release 18, ISO/IEC 19941, ETSI GR ZSM) to pinpoint deficiencies in existing approaches, especially the lack of unified, real‑time context handling. Second, it proposes a layered parametrisation framework that defines context variables at four levels: (i) User Equipment (UE) – device CPU/GPU capacity, display resolution, battery state; (ii) Edge Nodes – compute load, storage availability, proximity to UE; (iii) Fog/Cloud – scalable resources, batch‑processing capabilities; and (iv) Service Management – QoS/QoE targets, SLA constraints. Third, it develops a mathematical model that integrates these parameters into a mixed‑integer linear programming (MILP) formulation. Decision variables (x_{i}^{k}) indicate whether user (i)’s task is placed on layer (k). The objective function minimizes a weighted sum of latency, energy consumption, and operational cost while respecting constraints on maximum tolerable delay, energy budgets, and resource capacities.

A three‑stage context‑awareness loop is introduced: (a) real‑time monitoring of network metrics (RTT, jitter, throughput); (b) analysis of user interaction patterns (number of concurrent users, scene complexity, interaction frequency); and (c) prediction of rendering workload per device. These stages feed the optimizer, enabling proactive migration of tasks from edge to cloud (or vice‑versa) before QoS violations occur.

The authors validate the model through extensive simulations involving 100 users distributed across five geographic zones, comparing against a baseline static edge‑placement strategy. Results show a 30 % reduction in average end‑to‑end latency, a 22 % decrease in total energy consumption, and seamless scaling when user count spikes, confirming the model’s ability to maintain QoE under varying conditions. Moreover, the predictive migration component demonstrates early congestion avoidance, further improving performance.

Limitations are acknowledged: the study relies on simulated environments, omits detailed security and privacy considerations, and the AI‑driven prediction module requires richer training data for production deployment. Future work is outlined to include (i) implementation on real testbeds with heterogeneous hardware, (ii) development of lightweight, on‑device inference models for context estimation, (iii) standardised APIs for multi‑domain orchestration, and (iv) integration of green‑computing metrics to align with sustainability goals.

In summary, the paper delivers a rigorously formalised, context‑aware orchestration framework that bridges the gap between edge and cloud resources for multi‑user XR, demonstrating measurable gains in latency, energy efficiency, and service continuity, and paving the way for scalable, immersive XR experiences in forthcoming 5G/6G networks.