The digital era has a massive physical footprint. As Data Centers consume an ever-increasing share of global electricity, the “business as usual” approach to cloud orchestration is no longer viable. Therefore, in EXIGENCE we are moving beyond simple resource allocation toward a new paradigm, the Sustainability-Driven Orchestration.
By integrating Open Source MANO (OSM) and Kubernetes (K8s) with advanced telemetry from the eco-efficiency domain, we can build distributed systems that don’t just target performance but actively seek the lowest carbon footprint throughout its operation. Presented here is the proposed design for sustainable computing and autonomous systems for a typical Data Center domain.
The Orchestration Logic Center: Enhanced OSM & Eco-Control
At the heart of this architecture lies an evolution of OSM. To enable green operations, a new module called Eco-Control has been introduced. This module constitutes the sustainability logic center, managing two powerful mechanisms:
- Spatial Shifting: Eco-Control analyses the carbon intensity of different regional energy grids and the energy consumption. If a cluster in northern Europe is powered by wind while a southern one is burning coal, Eco-Control directs the Resource Orchestrator (RO) to deploy workloads where the energy is cleanest.
- Service Throttling: When the grid is stressed or carbon intensity peaks, Eco-Control can transition services between different variants. This means dynamically adjusting performance levels or even scaling back non-essential features to save power without dropping the service entirely.
The Infrastructure Core: Hardening Kubernetes for Sustainability
Kubernetes acts as a virtualised infrastructure manager (VIM), but a standard K8s cluster does not account for carbon emissions. At EXIGENCE, it has been enhanced with three key capabilities:
- Distributed Observability Agents: Every cluster runs a specialised agent that tracks hardware and software energy consumption in real-time, providing the granular data needed for sustainability orchestration at all levels.
- Kube-Scheduler with Green Score Plugin: Once OSM selects a cluster, the local k8s scheduler takes over. The Green Score Plugin is natively integrated in the K8s scheduling framework to provide an “eco scoring” of physical nodes based on the SCI (Software Carbon Intensity) specification, placing pods on the node that offers the best energy-to-performance ratio.
- Green HPA: The Horizontal Pod Autoscaler capabilities of K8s has been enhanced beyond tradition CPU and memory monitoring to be able to drive automated scaling based on sustainability metrics. Thus, enabling new use possibilities such as adjusting the number of pod replicas during periods of high carbon intensity in the energy mix.
Network Operators
The Foundation: Hierarchical & Multi-Domain Observability
The accuracy of the orchestration depends on the quality of the underlying metrics. This implementation uses a hierarchical observability layer featuring a centralised controller that aggregates telemetry from all virtual infrastructure managers and a series of distributed observability agents at each VIM. The system provides a Northbound API, which the green scheduler plugin and the Eco-Control module queries to make its “Spatial-Shifting” decisions, and an architecture based on a modular, cloud-native stack designed for multi-agent environments:
- Ingestion & Storage: OpenTelemetry and Thanos.
- Intelligence & Metrics: Alumet for energy profiling and other custom exporters for carbon intensity
- Semantic Abstraction: A system that offers shared semantics across Performance, Security, Trust, and Eco-efficiency domains.
Measuring What Matters: Software Carbon Intensity (SCI)
Network Operators
We utilise the SCI metric to drive improvement. Unlike traditional approaches that are easily improved by buying carbon credits, SCI enforces actual energy efficiency of the software and hardware. The equation to represent it is defined as the sum of operational emissions (the product of energy consumption and carbon intensity, E * I) and the embodied emissions of the physical hardware (M), normalised per functional unit of work (R).
Operational emissions represent the carbon emissions resulting from the use of software. Here, E is the energy consumed, and I is the carbon intensity based on location. Regarding embedded emissions, M corresponds to the fraction of carbon emissions, derived from the life cycle assessment (LCA) of a hardware component, that is allocated on an amortisation basis to the software running on it.
Core Capabilities and Operational Outcomes
The fundamental capabilities described here define the operational logic of the architecture approach, serving as the primary drivers for achieving deterministic sustainability outcomes.
- Energy-Aware Scheduling: The system profiles specific services to determine their optimal execution environment, ensuring placement on the most efficient hardware architectures available. This process aligns workload demands with physical resource capabilities to minimise wasted power.
- Carbon-Aware Placement: By analysing real time power grid intensity, this feature prioritises execution windows and geographic locations that utilise renewable energy. It effectively shifts demand to coincide with “green” periods, reducing the overall environmental impact of the distributed system.
- Multidimensional Metrics: The observability layer consolidates telemetry across the entire stack, from the cluster and node levels down to individual services and processes. This high-resolution data provides the foundation for precise resource management and deterministic system behaviour.
- Policy Based Control: A sophisticated governance layer enforces dynamic rules that balance operational performance with sustainability objectives.
In conclusion, by informing the orchestration with deep telemetry, ranging from anomaly detection in the security domain to carbon metrics in the Eco-efficiency domain, a system is created that is not only robust but responsible. We are no longer just managing containers; in EXIGENCE we are managing the environmental impact of the digital services.
Author
EVIDEN
Jesús Benedicto is a Systems Analyst/Software Architect at ATOS (EVIDEN) BDS Research & Development with over 18+ years of technological experience in software design and integration in industrial markets, with specialization in Big Data and Cloud-Edge-IoT technologies. Mainly focused on innovation in digital transformation and Smart automation in the manufacturing and telco sectors.