Technology Category
Machine Learning: Hardware-aware MLPlanning, Routing, and Scheduling: Plan Execution and MonitoringSearch and Optimization: Algorithm Configuration
Non-Intrusive MLOps-Driven Performance Intelligence in Software Data Planes
🤖 AI Summary
In NFV environments, resource contention among co-located VNFs on shared infrastructure causes significant service performance fluctuations—especially under gigabit-scale traffic—while conventional packet- or flow-level monitoring approaches suffer from prohibitive overhead and system-level constraints. This paper proposes a lightweight, non-intrusive online performance inference and adaptive optimization framework. It leverages hardware-assisted performance monitoring (e.g., Intel PCM) for low-overhead data-plane observation and integrates a traffic-agnostic, VNF-agnostic MLOps pipeline to enable runtime drift detection, bottleneck diagnosis, and autonomous tuning—without requiring traffic modeling or VNF modifications. Experimental evaluation across diverse NFV scenarios demonstrates that the framework accurately infers performance variations, reduces deployment overhead by up to 72%, and maintains service stability under high-throughput conditions.
Technology Category
Application Category
📝 Abstract
The last decade has witnessed the proliferation of network function virtualization (NFV) in the telco industry, thanks to its unparalleled flexibility, scalability, and cost-effectiveness. However, as the NFV infrastructure is shared by virtual network functions (VNFs), sporadic resource contentions are inevitable. Such contention makes it extremely challenging to guarantee the performance of the provisioned network services, especially in high-speed regimes (e.g., Gigabit Ethernet). Existing solutions typically rely on direct traffic analysis (e.g., packet- or flow-level measurements) to detect performance degradation and identify bottlenecks, which is not always applicable due to significant integration overhead and system-level constraints.
This paper complements existing solutions with a lightweight, non-intrusive framework for online performance inference and adaptation. Instead of direct data-plane collection, we reuse hardware features in the underlying NFV infrastructure, introducing negligible interference in the data plane. This framework can be integrated into existing NFV systems with minimal engineering effort and operates without the need for predefined traffic models or VNF-specific customization. Through comprehensive evaluation across diverse NFV scenarios, our Drift-Resilient and Self-Tuning (DRST) framework delivers accurate performance inference, runtime bottleneck diagnose, and automated adaptation under runtime drift, via a lightweight MLOps pipeline.
Problem
Research questions and friction points this paper is trying to address.
Detect performance degradation in NFV without direct traffic analysis
Address resource contention in shared NFV infrastructure
Enable lightweight online performance inference and adaptation
Innovation
Methods, ideas, or system contributions that make the work stand out.
Lightweight non-intrusive online performance framework
Reuses NFV hardware features for minimal interference
MLOps pipeline enables drift-resilient self-tuning adaptation
🔎 Similar Papers
No similar papers found.
