Measuring IT Carbon Footprint: What is the Current Status Actually?

Overview

Despite the EU Corporate Sustainability Reporting Directive pressing large enterprises to be more transparent about their GHG emissions, there are plenty of challenges ahead in measuring emissions from IT activities. For each area the paper reviews current tools (RAPL, IPMI, SPEC benchmarks, ElectricityMaps/WattTime data) and their limitations, and lists open research questions. Total IT carbon is defined as TotCarbon = (Es + Eo) × I + M, where Es is software energy, Eo server overhead energy, I the momentary energy-mix intensity and M embodied carbon; this is the GSF SCI formula without the functional unit R.

SDIA DEF model: Etot = Ecpu + Emem + EIO + Enet + beta_idle. Assuming TDP × N_CPU approximates CPU energy at full load, with fixed allocation factors alpha (summing to 1) and linear scaling, Ecpu = (Ucpu/Ucpu_max) × TDP × N_CPU and Emem = (Umem/Umem_max)(alpha...). The DEF model's fourth assumption takes idle consumption as zero; the paper itself raises this as an open question, noting that high idle power would favour powering off whole machines rather than spreading load.

Embodied carbon allocation: sum over objects of (manufacturing + repair/maintenance + end-of-life emissions) × (time consumed / usage-phase lifespan T); a dynamic version multiplies by a sharing factor f_c(tau, o_i) between 0 and 1 for each time step. Stated PUE limitations: it excludes IT equipment efficiency, meteorological conditions and waste-heat reuse, and is most meaningful at full IT capacity; complementary metrics listed are CUE, WUE, ERF, ITEE and OEM. Operational carbon over time is computed as PUE × sum over intervals of grid-mix intensity × energy, constrained by the granularity of energy-mix data (for example 30-minute updates); the paper discusses average versus marginal emission factors.

Limits: the paper contains no new measured data; its value lies in its accounting formulas and its discussion of scope, idle-time and shared-resource attribution problems. The DEF model is a rule of thumb for quick assessments, not a substitute for measurement.

Key points

1. Total IT carbon is TotCarbon = (Es + Eo) × I + M, where Es is software energy, Eo server overhead energy, I the momentary energy-mix intensity and M embodied carbon; the GSF SCI formula without the functional unit R.
2. SDIA DEF model: Etot = Ecpu + Emem + EIO + Enet + beta_idle; Ecpu = (Ucpu/Ucpu_max) × TDP × N_CPU.
3. The DEF model's fourth assumption takes idle consumption as zero; the paper notes this as an open question and that high idle power would favour powering off whole machines rather than spreading load.
4. Embodied carbon allocation: sum over objects of (manufacturing + repair/maintenance + end-of-life) × (time consumed / usage-phase lifespan T); a dynamic version multiplies by a sharing factor f_c between 0 and 1.
5. PUE limitations: it excludes IT equipment efficiency, meteorological conditions and waste-heat reuse, and is most meaningful at full IT capacity; complementary metrics are CUE, WUE, ERF, ITEE and OEM.
6. Operational carbon = PUE × sum over intervals of grid-mix intensity × energy, constrained by energy-mix data granularity (for example 30-minute updates); average versus marginal emission factors are discussed.

Using it in design

• Compute IT carbon as (Es + Eo) × I + M and allocate embodied carbon (M) for shared resources by time and resource share; define scope, idle-time and shared-resource attribution explicitly.
• For quick assessment use a rule of thumb based on CPU TDP and utilisation such as the DEF model, noting that idle consumption is assumed zero and that it does not replace measurement.
• Complement PUE with CUE, WUE, ERF and ITEE since it excludes IT efficiency, weather and heat reuse; document the choice of average versus marginal emission factor.

Who should read it

Sustainability and IT analysts choosing measurement methods for IT carbon-footprint accounting and Corporate Sustainability Reporting Directive compliance.

Document contents

1. 1 Introduction
2. 2 Four measurement areas (2.2 Existing Data and Tools, 2.5 Software energy consumption: 2.5.1 Empirical Estimations Of CPU Energy)
3. 3 Server overhead (PUE)
4. 4 Energy mix (4.2 Average vs Marginal)
5. 5 Embodied carbon (5.5 Potential Improvements)
6. 6 Conclusion

Related standards

• ISO/IEC 30134-2
• Green Software Foundation SCI specification
• EU Corporate Sustainability Reporting Directive

Read the document

Publication details

Authors

Tom Kennes

Publisher

arXiv preprint (cs.SE); Skyworkz / Sustainable Digital Infrastructure Alliance (SDIA)

Year

2023

Document number

arXiv:2306.10049

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.48550/arXiv.2306.10049

Licence

CC BY 4.0 — share with attribution

Licence evidence

arxiv.org

Citation