Embodied Water — Allocation Calculations
Embodied Water — Allocation Calculations
In short: Embodied Water introduced the shared-taxi-fare
idea — you're only charged for the stops you rode and the seats you took.
This page walks through choosing the four real-world inputs that pin down
those two fractions for a piece of shared hardware: TiR, EL, RR, and
ToR. If you have allocated embodied carbon under the SCI specification,
this will be a very familiar exercise — the mechanics are the same,
applied to a water total instead of a carbon total.
Time share: TS = TiR / EL
EL— expected operational lifetime. How long the hardware is expected to remain in service before replacement. Use the manufacturer's or operator's own expected-lifetime figure where available; where it isn't, a documented, disclosed assumption is acceptable (e.g. a typical server refresh cycle for the deployment context).TiR— time reserved for the software. How long, within that lifetime, this specific software system had the hardware reserved for it. For a long-running dedicated server, this may simply be the full reporting period. For hardware that is provisioned and torn down dynamically (e.g. autoscaled instances, serverless execution environments), this should reflect actual reservation time for the reporting period being assessed, not the hardware's full lifetime.
Resource share: RS = RR / ToR
ToR— total resources available. The hardware's total capacity in whatever resource dimension is most meaningful for the allocation — CPU cores, memory, or another consistent unit.RR— resources reserved for the software. The software's actual claim on that capacity. For a workload with a dedicated resource reservation (a fixed number of vCPUs, a fixed memory allocation), this is usually straightforward. For workloads that share resources dynamically with other tenants, this should reflect an average or peak reservation over the reporting period, disclosed as such.
Shared cloud hardware, specifically
Cloud hardware is almost always shared, which makes both TS and RS
non-trivial to pin down precisely. Some practical points:
- Where a cloud provider discloses per-instance resource entitlements
(e.g. "this instance type is allocated N vCPUs of a M-vCPU physical
host"), that ratio is a reasonable basis for
RS. - Where time-sharing is dynamic (e.g. serverless, autoscaling),
TSshould be computed from actual measured reservation time for the reporting period, not from a theoretical maximum. - Whatever data source and method is used for either share, disclose it — the specification's general disclosure requirements for allocation methods and assumptions apply here just as much as to the water totals themselves.
For where the underlying lifecycle water totals (TW_b) come from, see
Embodied Water — Datasets.
Please submit any comments you have here.