## Excerpt

The NADIKI system calculates the environmental impact of workloads based on real data — not on green electricity certificates. For CO₂ emissions, we use the actual CO₂ intensity of power generation at the data center location through the Electricity Maps API. For the manufacturing emissions of server hardware, we calculate a daily environmental impact budget using the Boavizta API, which is allocated to the workload proportionally to its usage. For the data center building and technology, we use an LCA from a reference data center if no LCA data is available.

## Content

The NADIKI Observer Architecture captures metrics from data centers and calculates environmental impact indicators for individual workloads. This article describes the methodology behind this calculation: What data sources are included, why we deliberately forgo market-based instruments like green electricity certificates, and how we calculate environmental impact for the workload.

Actual CO₂ Emissions Instead of Green Electricity Certificates

Many data centers cover their electricity consumption with certificates of origin or power purchase agreements with 100% green electricity. For an environmental impact calculation that is intended to support applications in operational decisions, these market-based instruments are unusable: The CO₂ value would often be zero and therefore not very meaningful — regardless of whether the power from the grid currently comes from coal, gas, or wind.

Instead, NADIKI uses the actual CO₂ intensity of power generation at the data center location. We use the API from Electricity Maps, which provides the real CO₂ content of power generation per country and region in real-time. We thank Electricity Maps for providing the API.

This approach provides information that is truly actionable for applications: A server can be shut down, a load shifted to another location — based on the real CO₂ content of the electricity generated in the region.

Embodied Emissions: Manufacturing Impact of the Hardware

In addition to operational emissions from electricity consumption, NADIKI considers the environmental impact of manufacturing server hardware — the so-called embodied emissions. For this, we use the API from Boavizta, which is based on data from the Fraunhofer Institute and established environmental databases.

For the calculation, we only need the server configuration: number of CPUs, power supplies, hard drives, storage capacity, and RAM. Boavizta provides an estimate of the environmental impact of the server's manufacturing.

We distribute this manufacturing impact over the configured lifespan of the server — in our standard configuration, five years, adjustable by the operator. This results in a daily environmental impact budget that is cumulatively assigned to the workload: The longer an application occupies the server resource, the more of the manufacturing impact is attributed to it.

An open question remains how to handle refurbished hardware: How is manufacturing impact distributed when a server has already gone through a first lifecycle? We address this topic in the further development of the calculation model.

Environmental Impact of the Data Center: Building and Infrastructure

In addition to server hardware and electricity consumption, the data center itself causes environmental impacts: emissions from the construction of the building and the manufacturing of electrical and mechanical infrastructure such as cooling, UPS systems, and power distribution. To also calculate this impact on the workload level, NADIKI needs an LCA model of the data center building and the infrastructure installed therein.

Ideally, the operator provides its own LCA, for example, from a BREEAM or LEED certification. However, for the pilot data centers connected to NADIKI, no such LCA was available. We use the LCA model of a data center in Sweden from the work of Felipe B. Oliveira as standard values: "Life Cycle Assessment of a High-Density Datacenter Cooling System: TeliaSonera's 'Green Room' Concept" (KTH Stockholm, 2012). The values are scaled to 1 kW of data center area and serve as a baseline that operators can replace with their own LCA data.

Cumulative Allocation at the Workload Level

The NADIKI system calculates the environmental impact of each workload cumulatively. Depending on the workload's utilization on the host system, its proportional environmental impact from three sources is allocated:

• Electricity Consumption: Proportional to server power consumption, multiplied by the actual CO₂ intensity on site (Electricity Maps).

• Hardware Manufacturing: Proportional to the daily embodied emissions budget of the server (Boavizta), proportional to the duration of use and reservation of server resources.

• Data Center Infrastructure: Proportional to building and infrastructure emissions, scaled over the workload's electricity consumption (LCA model).

These three components are cumulated over the application's lifetime. The NADIKI Registrar calculates the allocation automatically and provides the results through the query interface — the workload can query its current environmental impact at any time.