Please note: This PhD defence will take place in DC 1331.
Niloy Saha, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Raouf Boutaba
Next-generation mobile networks are undergoing a fundamental transformation to support a variety of services such as Augmented Reality (AR), vehicular automation, and smart cities, each with very distinct performance requirements. The primary enabler for this transformation is network slicing, which enables partitioning of the shared physical infrastructure into isolated logical networks, each tailored to the performance requirements of its target service through slice-specific Service Level Agreement (SLA) guarantees.
Realizing such network slices in practice depends critically on automated closed-loop management and orchestration mechanisms, where monitoring data drives analytical and control decisions. However, monitoring is a key bottleneck in achieving this goal: control logic can only optimize what it can observe, and observation is constrained by strict overhead, latency, and deployability limits. Existing monitoring solutions are domain-centric, statically configured, and not sufficiently expressive to capture the transient, end-to-end (e2e) performance signals that slice SLA assurance demands.
The overarching goal of this dissertation is to build the foundation for efficient, high-precision monitoring for network slices in 5G and beyond networks. Towards achieving this goal, we address three distinct but coupled challenges across the monitoring stack. First, we introduce Monarch, acloud-native architecture that abstracts slices as first-class monitored entities and enables direct computation of e2e slice Key Performance Indicators (KPIs) across heterogeneous domains. Next, we present SliceScope, a principled framework for dynamically allocating limited monitoring budget across slices with heterogeneous SLAs. Finally, we develop Kestrel, a detectability-driven, sketch-based telemetry mechanism for detecting and attributing transient Quality of Service (QoS) anomalies in shared user-plane resources without per-packet overhead. Together, these contributions demonstrate that efficiency and precision in slice monitoring are jointly achievable through cross-layer design spanning architectural abstractions, control plane intelligence, and data plane telemetry.