This study presents an empirical investigation into the energy consumption of discriminative and generative AI models within real-world MLOps pipelines. For discriminative models, we examine various architectures and hyperparameters during training and inference and identify energy-efficient practices. For generative AI, large language models (LLMs) are assessed, with a focus primarily on energy consumption across different model sizes and varying service requests. Our study employs software-based power measurements, ensuring ease of replication across diverse configurations, models, and datasets. We analyse multiple models and hardware setups to uncover correlations among various metrics, identifying key contributors to energy consumption. The results indicate that, for discriminative models, optimising architectures, hyperparameters, and hardware can significantly reduce energy consumption without sacrificing performance. For LLMs, energy efficiency depends on balancing model size, reasoning complexity, and request-handling capacity, as larger models do not necessarily consume more energy when utilisation remains low. This analysis provides practical guidelines for designing green and sustainable ML operations, emphasising energy consumption and carbon-footprint reductions while maintaining performance. This paper can serve as a benchmark for accurately estimating total energy use across different types of AI models.

The scope of the Sixth-Generation Self-Organized Networks (6G-SON) advances its predecessor's capability towards agility, flexibility, and adaptability. On-demand overlay networking technologies have shown a prominent maturity while coping with the rising complexity and scale of enterprise, service provider, and data centre networks. The Software-Defined Networking paradigm has recently offered Model Driven Programmability, minimizing network management complexity through automation and orchestration. However, leveraging Machine Learning-driven network optimization, a.k.a. Knowledge-Defined Networking (KDN), has still been a domain of interest for the Network Softwarization research community. In this article, we propose Intelligent Routing as a Service (iRaaS) architecture as an application layer cognitive routing framework for KDNs. iRaaS offers routing logic customization (i.e., customizing metric function and path-finding algorithm) and provides an option to include heuristic parameters from trained models as a part of the metric calculation. iRaaS sits on the application plane above the knowledge plane in a KDN stack, thus providing platform- and vendor-agnostic coupling with existing network infrastructures. This article covers the scope of iRaaS by using reliability as a heuristic for standard path-discovery algorithms, e.g., Shortest Path First (SPF) and Diffusion Update algorithm (DUAL), along with the architectural specification. We validate our approach through a Proof-of-Concept deployment.

GitOps has emerged as a foundational paradigm for managing cloud-native infrastructures by enabling declarative configuration, version-controlled state, and automated reconciliation between intents and runtime deployments. Despite its widespread adoption, the performance and scalability of GitOps tools in Intent-Based Networking (IBN) scenarios are insufficiently evaluated. This paper presents a reproducible, metric-driven benchmarking, assessing the latency and resource overheads of three widely used GitOps operators: Argo CD, Flux CD, and ConfigSync. We conduct controlled experiments under both single- and multi-intent scenarios, capturing key performance indicators such as latency and resource consumption. Our results highlight trade-offs between the tools in terms of determinism, resource efficiency, and responsiveness. We further investigate a realistic orchestration scenario, using Nephio as our orchestrator, to quantify the processing latency and overhead in declarative end-to-end deployment pipelines. Our findings can offer valuable insights for tool selection and optimisation in future autonomous network orchestration systems.