A tier above PUE: the global KPI in ETSI EN 305 200-3-1

The contract for running a data centre asks for quarterly PUE under a contractual threshold. The operator delivers it, on target. Whoever pays the electricity bill — or has to declare the site’s environmental footprint to a client, a partner, an authority — asks the next question: does that figure include the heat the site sells back to the neighbourhood? Does it include the electricity from a photovoltaic plant shared with three other sites in the same group? And who measured it, at which point, with what declared margin? PUE answers none of those three questions: it was never written to. There is a public technical document that puts all three into the same number — provided each one declares where and how it was measured.

ETSI EN 305 200-3-1, and the family it belongs to

We downloaded ETSI EN 305 200-3-1 V1.1.1 (2018-02), Energy management; Operational infrastructures; Global KPIs; Part 3: ICT Sites; Sub-part 1: DCEM, from ETSI’s own listing and read it in full: the only published edition, with a draft rejected in March 2017 (as ES 205 200-3) and publication on 20 February 2018. DCEM stands for Data processing and Communications Energy Management. It is the first and only published sub-part of the -3 series, a simplified implementation of the full KPI that -2-1 (ICT sites, V1.1.1 2018-02) defines under the common framework set out in -1 (general requirements, V1.1.1 2018-07). The text says so of itself, in clause 1: “The present document specifies the requirements for a Global KPI for energy management” — simplified from the fuller -2-1 KPI.

What it adds, beyond the ratio of two energies

PUE — ITU-T L.1302, which we have already written about, fixes its measurement points and frequency — stays a ratio of two energies. KPI_DCEM includes it, but only as one ingredient among four: the standard says so in one line, in clause 3.1, defining energy management as a “combination of reduced energy consumption and increased task effectiveness, re-use of energy and use of renewable energy” — four objectives, not one.

KPI_DCEM combines two indicators. The first, KPI_EC, is absolute consumption in MWh, banded into seven classes from XXS to XXL (DCG): it says how big the site is, nothing about how well it runs. The second, KPI_EP, is energy performance, and this is where PUE enters as an ingredient: KPI_EP = KPI_TE × (1 − W_REUSE × KPI_REUSE) × (1 − W_REN × KPI_REN). KPI_TE — which the standard itself notes approximates the power usage effectiveness of EN 50600-4-2:2016 (Power Usage Effectiveness), a paid standard known here only from its catalogue listing, not read — is defined in clause 3.1 as “measure of the work done (as a result of design and/or operational procedures) for a given amount of energy consumed”, with a declared ideal value of 1. The two weighted subtractions that correct it are heat re-used outside the site (KPI_REUSE) and renewable energy (KPI_REN). The result, banded into seven classes from A to G (DCCLASS), gives one ratio of scale and one of merit: never a single figure that buries one inside the other.

The boundary runs through more hands than a specification usually accounts for

The standard draws the site’s energy perimeter with two schemes — electrical energy, thermal flows — and for a mixed-use premises it adds the points that separate ICT consumption from the rest of the building, on top of the four required for a dedicated site. The document is explicit about what stays out of scope: it “does not address other layers such as performance of ICT equipment itself, performance of usage of available processing power, and layers related to final service delivered”. It measures the site’s infrastructure, not the efficiency of whatever workload runs on top of it — that boundary is worth holding onto before anyone reads a KPI_DCEM figure as a verdict on the equipment inside.

Each measurement point sits under a different party’s responsibility: grid-delivered energy in the supplier’s meter, local generation in the site operator’s own plant, re-used heat often metered inside a third party’s district heating network. The standard does not leave the point to good faith: it classifies each one, and assigns a declared penalty to every class. For energy consumption, metering at the transformer’s input counts as zero; at its output, 1.5%; at the input to the primary distribution switchgear, 2%. For the ICT equipment load — the denominator of KPI_TE — the socket feeding it directly counts as zero; the secondary distribution switchgear, sometimes called the departmental PDU, 1%; the UPS output, 2%. This is not an appendix footnote: it is the difference between a declared KPI and a verifiable one, and the standard itself puts a number on it.

The discipline of time, and the weighting that does not get renegotiated

The default assessment period, T_KPI, is 365 days; the four objective KPIs may be measured on windows that differ by no more than 7 days (Δt) from one another, and the repeat interval, T_REPEAT, falls between 7 days and one month. On weightings the standard leaves no room for manoeuvre: “When applied to a group of ICT sites under common governance, the same weighting shall apply to all ICT sites” — a single W_REUSE and a single W_REN for the whole portfolio, not a more forgiving one for whichever site needs to show better numbers this quarter. Grouping itself carries its own condition: “The selection of ICT sites to be included in groups shall reflect a common purpose”, not a merge that dilutes a lagging site inside the average of a leading one. Weightings and bands stay free between 0 and 1: the informative annex reports only example values, identical for CRIP and eG4U (W_REUSE 1.0, W_REN 0.8).

What to write into a specification

The declared measurement point for each objective KPI, together with the penalty class the standard assigns it — not a generic measured to standard with no stated address.
W_REUSE and W_REN written down before the first measurement, held fixed for the whole comparison period rather than adjusted quarter to quarter.
The perimeter of any grouped sites, where a contract covers more than one, with the common purpose that holds the group together.
The full report, not just the final class: the standard asks for KPI_EC and DCCLASS at minimum, but for trend analysis to work it also requires KPI_EP, KPI_TE, KPI_REUSE and KPI_REN, each with its own weighting.
Who holds each data point: grid supplier, site operator, a third party’s district heating network — with the link between those data sources written down, not assumed.

See the service · Talk to an engineer

At acceptance, check that the declared measurement point actually exists where the paperwork says it does, that the applied penalty matches the real class rather than the most favourable one, and that this period’s weightings match the previous period’s — not a silent adjustment to flatter the trend line.

The point

KPI_DCEM does not replace either L.1302 or L.1322: the first lends it the KPI_TE ingredient, the second sits underneath, explaining why that ingredient reads the way it does. KPI_DCEM sits a tier above both, and adds two things neither PUE nor the containment indices touch: how much energy the site hands back beyond its own boundary, and how much of what it draws comes from sources that never pass through the shared grid — one banded figure, comparable across sites of very different scale, provided the specification demands the path that backs it up, not the final figure on its own.

For a data centre, with CSIDIA, the group’s other company, the energy measurement points — grid delivery, local generation, re-used heat, ICT equipment load — stop being four separate archives and become a single model, on which an AI tracks the drift of KPI_TE, KPI_REUSE and KPI_REN between periods, before the final class quietly changes letter with no written cause. Within the client’s own perimeter: on-premises on autonomous machines, or dedicated cloud with a data centre in Italy, always with shared management between the two teams.

Do you need to declare a site’s KPI_DCEM, or check one a supplier has already declared? Talk to an engineer: the site visit is free, and the measurement point gets written into the specification before the first report is due, not after.