Outcome-aware · Regional · Compute · Allocation

ICT4S 2026 · Bern · ConverStation

The greenest
schedule isn't
always green.

Workload shifting beyond carbon intensity — and who pays the bill you can't see in a CO₂ number.

Geerd-Dietger Hoffmann · didi@green-coding.io
Verena Majuntke · verena.majuntke@htw-berlin.de
Green Coding Solutions · Uni Potsdam · HTW Berlin

Around the table

Where would you place a workload?

The pitch every cloud team has heard

"Run the job where and when the grid is cleanest."

Carbon intensity (gCO₂e/kWh) swings by region and hour. So we move delay-tolerant work to the low-carbon slot. Simple, measurable, and almost everyone optimizes for this one number.

→ What does that one number quietly leave out?

Around the table

You're the scheduler. One Django CI run.
Pick the greenest data centre.

Hour 05:00 UTC · 1 Jul

The hidden ledger

Same hour. Same job. Now you can see
the local burdens each choice triggers.

Hour 05:00 UTC · 1 Jul

And CO₂ was only the start

There are many other impacts.

What computing actually touches

Electricity
Water
Land usage
Rare earths
Waste management
Air pollution
Human

New factors emerging

Noise
Light
Refrigerant leakage
Metal for construction
Social / society

CO₂ vs. the others

How should we rate which one is more important?

Now Future

Noise

Air pollution

Water

CO₂

Nuclear waste

ORCA's core move · RQ1

Tag every impact with the zone where it's actually felt.

Globalclimate, no local exposure

CO₂ · CH₄ · N₂O

Energy Productionat the power plant

SO₂ · NOₓ · CO · NM-VOC

Data Centreat the facility & its city

Water stress · NO₂ · PM2.5 · price

How a choice gets scored

One weighted score,
many dimensions.

Each impact e gets a burden, normalized so liters and grams can sit side by side, then weighted by how much you care about it.

THE GOVERNANCE KNOB

A cap can forbid any choice that pushes a zone past its limit — even if it wins on CO₂.

Iₑ(d,τ) = Σ_i u(i)·g_d,e(τ+iΔ) + T_e(o,d)

Σ u·g execution: energy curve × intensity
T_e transfer: burden to move the data

↓ feeds into

J(d,τ) = Σ ωₑ · Iₑ(d,τ) / sₑ

ωₑ your weight per impact · sₑ normalizer
J pick the smallest

Now you decide · the part I actually want from you

You're tonight's on-call engineer. The CI batch is cheapest-carbon in Tokyo — but Tokyo's basin is water-stressed and Frankfurt's isn't. Frankfurt costs ~25 g more CO₂ per run. You run it 10,000× tonight.

One call. No "it depends." Pick a corner.

The turn — once you've committed

Who should be allowed to set that weight? The engineer, the company, the host country, or the people downwind?

Thank you — and a favour

Next paper: why don't people actually migrate their workloads?

I genuinely need this table's help. Tell me the real reason your jobs stay put.

didi@green-coding.io · verena.majuntke@htw-berlin.de
linkedin.com/in/dietgerhoffmann · linkedin.com/in/verena-majuntke

Code & data: AGPL, fully public

Scan to explore

shift.sensationtech.com

The greenestschedule isn'talways green.