Predictive Hierarchical Orchestration for Multi-Domain Network Slicing

METIS: A Declarative Slice Orchestrator for Application-Centric 5G/6G Networks

Network slicing is the cornerstone of application-aware 5G and 6G networks, yet dynamic lifecycle management of network slice instances with coordinated quality-of-service enforcement across the radio access network and core network remains unresolved. Existing orchestrators rely on network-centric data models, imperative workflows, and static slice templates, while O-RAN addresses radio-side slice control independently of 3GPP core-side control, leaving slice-level quality-of-service enforcement uncoordinated across domains. This paper introduces METIS, a declarative slice orchestrator that manages the Day-0/1/2 lifecycle of network slice instances through cascaded reconciliation loops. METIS defines an application-centric data model for service profiles, enabling customers to describe the semantics and quality-of-experience requirements of their applications. From these, METIS derives 3GPP-aligned slice profiles via hierarchical aggregation following the 5G quality-of-service model, eliminating static templates, and jointly coordinates O-RAN and 3GPP slicing for slice instantiation and enforcement. Our central finding is a structural asymmetry in end-to-end slice control: downlink traffic can be shaped at the core before reaching the radio access network, but uplink leaves the user equipment unregulated, so core-only slicing cannot reliably satisfy uplink service-level agreements - radio-side enforcement is necessary, not merely complementary. Evaluated on a 5G cloud-native testbed in a campus-event scenario, METIS completes slice creation, update, upgrade, and deletion within 22.4, 5.1, 52.2, and 32.1 seconds, respectively; sustains full service-level-agreement satisfaction under concurrent multi-slice overload; scales to 63 slice instances across nine zones consuming under 0.03 processor cores total; and recovers slices from injected failures across four levels in under 19 seconds.

SLA-AWare RAN Slicing Via Online Meta-Learning

Real-time inter-slice resource allocation in the Radio Access Network (RAN) is a critical control function in 5G and emerging 6G networks, where the scheduler in the Distributed Unit (DU) dynamically allocates physical resources, namely Physical Resource Blocks (PRBs), to different network slices to meet their diverse Quality of Service (QoS) requirements. To address the need for faster and more flexible radio resource management, and inspired by recent efforts to extend the O-RAN architecture with a real-time controller, we investigate slice-level PRB allocation through the lens of online learning. We formulate inter-slice scheduling as a dynamic decision problem and develop a system model that captures per-slice Service Level Agreement (SLA) requirements and throughput variations over configurable time windows, without assuming future channel knowledge. Our scheduling solution is implemented as a real-time RAN control application, in line with the O-RAN proposition for dApps that are programmable and distributed software components for fine-grained control in O-RAN DUs (O-DUs) and Centralized Units (O-CUs). The proposed approach adapts inter-slice radio resource allocations based on telemetry, with low computational complexity. Experimental results show sublinear dynamic regret, up to 85% fewer SLA violations than static baselines, and submillisecond amortized control overhead. Overall, these findings highlight dynamic-benchmark online control as a practical mechanism for real-time, SLA-aware slicing in O-RAN.

Comprehensive Review of Optimization Techniques for User-Centric Distributed Network Slicing in 5G Networks

A QoE-aware framework for Multi-Access Edge Computing-enabled Open Radio Access Network (O-RAN) architectures, combining a graph attention network (GAT) encoder, distributed multi-agent DRL, and privacy-preserving FL, while transitioning control from Quality of Service (QoS) to QoE metrics is proposed.

SLO-Aware Graph Forecasting for Intelligent Autoscaling in API and Microservice Backends

Autoscaling microservice backends is challenging due to complex inter-service dependencies, bursty workloads, and cascading latency amplification along call chains. Existing approaches either react to threshold breaches after SLO violations occur, or predict demand per service in isolation. We propose SAGF (SLO-Aware Graph Forecasting), an integrated prediction-and-control framework for proactive autoscaling. SAGF encodes six per-service metrics on a spatio-temporal call graph, predicts per-service SLO risk via GCN with temporal gating, and feeds these scores to a constrained RL controller for dependency-aware scaling. On the Online Boutique benchmark with Kubernetes and Istio, SAGF reduces SLO violations by 41% and excess overprovision by 48% on average compared to the strongest baseline across four workload patterns.

Programmable B5G Infrastructure for Vertical-Aware Slice Orchestration: Teleoperation in Dynamic 5G Environments

Two-level Network Bandwidth Allocation for Multi-tenant AI Clouds in Kubernetes

Compute and memory resources in cloud environments are strictly managed and isolated by the control plane; in contrast, network resources lack equivalent management and isolation mechanisms. This best-effort treatment of networking leads to significant challenges for modern AI workloads, which have diverse and bandwidth-intensive communication patterns. Without fine-grained network resource control, these workloads suffer from interference, unpredictable throughput, and suboptimal cluster utilization. To address these issues, this paper demonstrates how network bandwidth can be elevated to a first-class, schedulable, and enforceable resource within Kubernetes, the de facto standard for cloud-native orchestration. We introduce a new scheduling capability that models network interfaces as allocatable resources and regulates bandwidth sharing through the Dynamic Resource Allocation (DRA) framework, with enforcement implemented using the Hierarchical Token Bucket (HTB) mechanism. We evaluate the system using multitenant AI workloads derived from real-world communication characteristics with a simulation-based approach and validate the proposed enforcement strategy in a real cluster. Results show that the proposed two-level bandwidth allocation improves tenant performance predictability and satisfaction while maintaining packed cluster utilization.