Skip to main content

You are viewing a project that is currently in draft state for the Standards Working Group in the Green Software Foundation. This project should not be considered finished or officially supported in any way by the Green Software Foundation or its members.

Functional Unit (R)

Functional Unit (R)

In short: every SWI number is a rate — water per something, not a raw total. R is that something. Get it wrong and the number stops being comparable to anything, even to a different measurement of the same system.

R is the reference unit an SWI value is expressed per: an API call, a user, an AI training run, or whatever quantified performance characteristic best describes how the software system actually scales. Every SWI value is a rate — water per unit of R — so choosing R well is foundational to the result being meaningful at all.

Choosing R​

The specification's own guidance is simple to state and easy to get wrong in practice: the functional unit should match how the software scales. A batch training job scales with training runs, not with wall-clock time; a user-facing API scales with requests or active users, not with server count. Pick the unit that tracks the thing that actually drives the software's resource use, not the unit that happens to be easiest to measure.

This is the same discipline the SCI specification applies to its own functional unit — if you've chosen R for an SCI assessment of the same system, that choice is very likely to transfer directly to its SWI assessment.

Every component must scale to the same R​

All elements of the SWI calculation — direct operational water, indirect operational water, and embodied water — must be expressed against the same functional unit before they're combined. It's not enough for the final result to be labelled "per R"; each underlying component has to actually have been measured or estimated on a basis that scales consistently with R.

Combining components with different functional units​

Sometimes an aggregate SWI is composed from sub-components that were naturally measured against different functional units — one service scales per request, another per user session. Where that happens, the specification requires:

  • All components to share the same functional unit, water accounting basis, adjustment status (unadjusted or stress-adjusted), system boundary, and a compatible characterization methodology before they can be summed.
  • Any component using a different functional unit to be converted to the aggregate's functional unit before aggregation — with the conversion factor and its assumptions fully disclosed.
  • Consumption and withdrawal results never aggregated with one another, and unadjusted and stress-adjusted results never aggregated with one another — this is the same non-substitution principle covered on the Water Accounting Basis page, extended to the point of aggregation.

Please submit any comments you have here.