Green cloud and sustainability
Green cloud and sustainability: energy- and carbon-aware design, metrics, techniques (virtualization, placement, cooling, renewables) and how to compute savings for cloud data centres.
Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.
Why it matters
Two or three sentences: Cloud engineers design and run data centres and cloud platforms. Making clouds "green" (energy- and carbon-efficient) reduces operational cost, meets regulation and customer sustainability targets, and is required when planning capacity, SLAs and disaster recovery.
Key ideas
Carbon footprint — total greenhouse gas emissions (typically expressed as CO2 equivalent, kgCO2e) associated with an activity. For a data centre, it includes Scope 2 electricity (location- or market-based), Scope 1 on-site fuels and refrigerant leakage (e.g., generators, chillers), and Scope 3 embodied emissions from manufacturing, transport, and end-of-life.
Power Usage Effectiveness (PUE) — defined as: PUE = (Total facility energy) / (IT equipment energy). This is an exact definition used by industry. Assumption: IT equipment energy > 0. Lower PUE means better facility-level energy efficiency; theoretical minimum is 1.0 (all energy used by IT equipment only). Data Center infrastructure Efficiency (DCiE) = 1/PUE (often expressed as a percentage).
- To draw a diagram for PUE: draw a rectangular box labelled "Data centre facility". Inside it draw a smaller box labelled "IT equipment". Draw arrows entering the big box labelled "Total facility energy" and split it: one arrow into IT equipment labelled "E_IT" and other arrows to UPS, cooling, lighting, power distribution labelled with their energy contributions. Label PUE = E_total / E_IT.
Carbon Usage Effectiveness (CUE) — defined as: CUE = (Total facility CO2e emissions) / (IT equipment energy). This links energy efficiency to carbon intensity of the electricity used. Assumption: same time period for numerator and denominator (annual is common). If emissions are entirely from electricity, CUE = PUE × grid emission factor (kgCO2e/kWh).
Water Usage Effectiveness (WUE) — defined as: WUE = (Annual water usage for facility) / (IT equipment energy). Useful where cooling uses water (evaporative, cooling towers).
Embodied energy (or embodied carbon) — the energy (or CO2) consumed/emitted in manufacture, transport, and disposal of equipment. It must be added to operational energy to compute lifecycle impact. Allocate embodied carbon over useful life (e.g., kgCO2e/year) or per service (e.g., per kWh by dividing by lifetime energy use) when comparing options.
Energy-proportional computing — a system property where power consumption scales closely with workload. Ideal: zero power at zero load. Real servers consume significant idle power; improving proportionality reduces energy waste at low utilisation.
Server consolidation (via virtualization) — running multiple workloads on fewer physical servers to increase average utilisation and reduce total energy. Assumptions: consolidated workloads fit on fewer servers without violating SLAs; virtualization overhead is small compared to idle power saved.
Workload placement and scheduling — placing and timing workloads to exploit cheaper or cleaner electricity (time-shifting), or to concentrate workloads in the most efficient data centres. Assumption: workloads are flexible in time/placement or have geo-flexible SLAs.
Exact limits and trade-offs: reducing PUE from 1.6 to 1.2 saves facility overhead energy but cannot reduce IT energy. Improving workload efficiency lowers IT energy directly. Always separate facility-level metrics (PUE, WUE) from IT-level improvements (energy proportionality, code efficiency).
Formulas
PUE = E_total / E_IT — E_total: total facility energy (J or kWh), E_IT: energy consumed by IT equipment (J or kWh). Applies to same time period.
CUE = CO2_total / E_IT — CO2_total: total CO2e emissions from facility energy (kgCO2e), E_IT as above (kWh). Applies per period (annual common). Units: kgCO2e/kWh.
WUE = Water_total / E_IT — Water_total in litres, E_IT in kWh; gives litres per kWh.
E = P_avg · t — E: energy (J or kWh), P_avg: average power (W), t: time (s or h). Use consistent units (W·s = J, W·h = Wh).
E_saved = (P_before - P_after) · t — power difference times time; used for consolidation savings.
CO2e = E · f — CO2e emissions (kgCO2e) = energy (kWh) · grid/carbon emission factor f (kgCO2e/kWh). Requires correct local f.
Notes: SI units: power (W), energy (J) or commonly kWh, CO2e in kgCO2e. Always state the time period (day, year) used.
Worked examples
- PUE and annual CO2 for a small facility
- Given: IT energy = 500 kWh per day. Total facility energy = 800 kWh per day.
- Step 1: Compute daily PUE:
PUE = 800 / 500 = 1.6. - Step 2: Annual energies: E_IT_year = 500 · 365 = 182,500 kWh; E_total_year = 800 · 365 = 292,000 kWh.
- Step 3: If grid emission factor f = 0.70 kgCO2e/kWh, annual CO2e = 292,000 · 0.70 = 204,400 kgCO2e = 204.4 tCO2e.
- Final answers: PUE = 1.6; annual facility energy = 292,000 kWh; annual CO2e ≈ 204,400 kgCO2e.
(If your local grid factor differs, replace 0.70 with the correct kgCO2e/kWh.)
- Energy saved by server consolidation (virtualization)
- Given: 100 servers each average power 250 W. After consolidation, 25 servers average 400 W. Calculate annual energy saved and CO2 saved (use f = 0.70 kgCO2/kWh).
- Step 1: Before consolidation power = 100 · 250 W = 25,000 W = 25 kW.
- Step 2: After consolidation power = 25 · 400 W = 10,000 W = 10 kW.
- Step 3: Power saved = 25 kW − 10 kW = 15 kW.
- Step 4: Annual energy saved = 15 kW · 24 h/day · 365 day = 131,400 kWh.
- Step 5: Annual CO2e saved = 131,400 kWh · 0.70 kgCO2e/kWh = 91,980 kgCO2e ≈ 91.98 tCO2e.
- Final answers: Energy saved = 131,400 kWh/year; CO2e saved ≈ 91,980 kgCO2e/year.
- Model answer (theory) — List and explain measures to make a cloud data centre sustainable (short essay style for exams)
- Start with definitions: define PUE and carbon footprint. Organise the answer into "Design", "Run-time", "Procurement & lifecycle", and "Policy".
- Design: choose efficient chillers, free cooling (using outside air when possible), efficient UPS and PDUs, hot-aisle/cold-aisle containment. Explain that these reduce E_total in PUE.
- Run-time: use virtualization for consolidation, autoscaling to turn off idle servers, workload placement across regions/time to exploit renewable supply or low-carbon grids, and demand response to reduce peak loads.
- Procurement & lifecycle: buy energy-efficient servers, prefer CPUs with better performance-per-watt, consider embodied carbon in purchase decisions, plan reuse and recycling.
- Policy & measurement: monitor PUE, CUE, and WUE; report annually; use green power purchase agreements (PPAs) or renewable energy certificates (RECs) to match consumption.
- Conclude by noting trade-offs: e.g., concentration to one site improves efficiency but increases geographic risk; life-cycle choices may increase capital cost but reduce lifetime emissions.
- This structure gains full marks: definitions, organised points, link to metrics, and mention of trade-offs.
Common mistakes
- Forgetting embodied energy when reporting lifecycle emissions.
- Assuming energy scales linearly with utilisation; many servers have high idle power.
- Using inconsistent time periods (e.g., daily E_IT with annual CO2 without conversion).
- Mixing units (using J and kWh without conversion). 1 kWh = 3.6·10^6 J.
For your exam
Kinds of questions: definitions (PUE, CUE, embodied energy), short notes (techniques: virtualization, free cooling), derivations (PUE and its bounds; energy saved formulas), numerical problems (compute PUE, savings, CO2), and diagrams (data centre showing IT vs facility energy).
What a full-marks answer includes:
- For definitions: precise formula and units, and at least one simple implication (e.g., PUE ≥ 1).
- For derivations: clear step-by-step algebra, stated assumptions (time period, constant power or average), and units.
- For numericals: list given data, formulas used, step-by-step calculation, final value with units and reasonable rounding.
- For diagrams: labelled boxes for IT equipment, cooling, power distribution, arrows for energy flows, labels for numerator/denominator of PUE.
Quick check
- What does PUE measure?
- Give the formula to compute energy from average power and time.
- Name two operational techniques that reduce IT equipment energy.
Answers: PUE measures facility energy divided by IT equipment energy; E = P_avg · t; virtualization (server consolidation) and autoscaling (turning off idle servers).
Finished this topic? Mark it so your progress, study plan and readiness keep up.
Stuck on something here?