Worked Examples
Worked Examples
This page walks one small, concrete scenario through every term in the SWI
formula, the same way the SCI Guidance Quick Guide walks an
e2-standard-4 VM through SCI = (E×I) + M per R. All numbers below are
illustrative placeholders, clearly labelled as such — they demonstrate
the mechanics of the calculation, not published water-intensity factors.
Do not treat any figure on this page as a real measurement.
The scenario
A containerised API workload runs on a single virtual machine in a cloud region, handling a steady stream of requests. We want its SWI, on both a consumption and a withdrawal basis, for one month of operation.
Step 1 — Bound
Following the spec's Bound step: the software boundary for this example is the containerised application and the virtual machine it runs on, for one calendar month. We exclude the client devices making requests and any downstream services not controlled by this workload.
Step 2 — Scale
Following Scale: this API's costs scale with traffic, so we choose
R = 1,000 API requests as the functional unit. (An illustrative
example only — a real assessment would justify this choice against how the
specific workload actually scales, per the
Functional Unit guidance.)
Step 3 — Define
Following Define: for this illustrative walkthrough we use modelled estimates throughout — energy from the cloud provider's usage dashboard, WUE and EWIF from published facility/regional figures (illustrative values, not sourced from a specific real facility for this example), embodied water from a time-share/resource-share allocation of the VM's underlying hardware.
Step 4 — Quantify
Inputs (illustrative placeholders):
| Input | Placeholder value |
|---|---|
Energy consumed, E (for 20,000 requests this month) | 5 kWh |
Facility WUE | 0.3 L/kWh (illustrative) |
EWIF_c (consumption basis) | 1.8 L/kWh (illustrative) |
EWIF_w (withdrawal basis) | 4.2 L/kWh (illustrative) |
Embodied water allocated to this VM, W_M,c / W_M,w | 0.4 L / 0.9 L (illustrative, already time-share/resource-share allocated) |
Direct operational water (consumption basis):
W_O,c^dir = E × WUE = 5 × 0.3 = 1.5 L
(No standardised withdrawal equivalent to WUE exists — see Direct Operational Water for why direct operational withdrawal is left unquantified in this example.)
Indirect operational water:
W_O,c^ind = E × EWIF_c = 5 × 1.8 = 9.0 L
W_O,w^ind = E × EWIF_w = 5 × 4.2 = 21.0 L
Totals for 20,000 requests this month:
| Consumption | Withdrawal | |
|---|---|---|
W_O^dir | 1.5 L | not quantified (data gap) |
W_O^ind | 9.0 L | 21.0 L |
W_M | 0.4 L | 0.9 L |
W_b | 10.9 L | 21.9 L (indirect + embodied only) |
Step 5 — Report, unadjusted SWI, scaled to R = 1,000 requests:
SWI_c = 10.9 / 20 = 0.545 L per 1,000 requests
SWI_w = 21.9 / 20 = 1.095 L per 1,000 requests
A full report would also apply characterization factors (CF_c,i,
CF_w,i) for the region where the electricity was generated, per
Stress Characterisation, to produce SWI_c^adj and
SWI_w^adj — omitted here since this walkthrough is illustrating the
unadjusted mechanics only.
What this example is not
A single-VM, single-month illustration like this is deliberately small. A
more rigorous worked example — stress-testing the formula across multiple
power generation units feeding a single data centre, with real
characterization data — has been proposed within the SSWG and is expected
to be contributed as a full page or case study once drafted. It is not
fabricated here; when it lands, it will supersede this page as the primary
worked example and this page will likely narrow to the introductory role
the SCI Quick Guide's e2-standard-4 example plays. The
Workday AI worked example
in the SCI-for-AI guidance is the precedent for the depth expected of that
fuller example.
Please submit any comments you have here.