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.

Server-Side Emissions (O_server)

In short: O_server is the carbon from the electricity used by every server-side system that builds and delivers your application's responses, including idle capacity you have reserved and the data centre's own overheads such as cooling. It stops at the point where data leaves the data centre for the network.

Shape: (compute energy × data-centre overhead factor) × grid carbon intensity = server emissions.

Worked: 1,000 kWh of compute × a PUE of 1.2 = 1,200 kWh; 1,200 kWh × 250 gCO2eq/kWh = 300 kg CO2eq for the month. (illustrative placeholder, not a real measurement)

Precise notation:

O_server = E_server × I_server
E_server = E_compute × PUE

where E_server is energy in kWh per functional unit (or per period, divided by R later) and I_server is location-based carbon intensity in gCO2eq/kWh.

What counts as server energy​

The original draft listed the infrastructure that E_server covers:

  • origin and application servers;
  • database servers and caching infrastructure;
  • serverless function execution environments;
  • edge computing nodes (CDN edge servers, edge functions) whose energy is measured as facility energy;
  • supporting infrastructure (load balancers, firewalls, monitoring).

Clause 7.2 requires the energy of all hardware reserved or provisioned for the application, including idle capacity. A server sized for peak traffic that sits mostly idle still counts in full; this is what makes right-sizing show up in the score. Monitoring, observability, redundancy and failover infrastructure are included where material, and Clause 5.2 requires any exclusion of them to be disclosed with its rationale.

Where server energy stops and network energy starts, including how to treat CDN and edge energy, is set out once, on Network Emissions.

Data-centre overhead: PUE​

Power usage effectiveness (PUE) is the ratio of a facility's total energy to the energy delivered to its IT equipment. Clause 7.2 requires facility overhead to be included; multiplying compute energy by PUE is the usual way to do it:

E_server = E_compute × PUE

Use the facility's or provider's published PUE for the period where possible and disclose its source. If a provider's energy figure already includes facility overhead, do not apply PUE a second time.

Shared infrastructure​

On multi-tenant infrastructure (shared hosting, public cloud), Clause 7.2 requires energy to be allocated by resource share and the method disclosed. A common approach is to allocate by the resources reserved for the application (vCPUs, memory, storage) as a share of the host's total capacity:

E_app = E_host × (resources reserved for the application ÷ total host resources)

State which resource you allocated by, and why it fits the workload.

Several grid regions​

Where infrastructure runs in several regions, Clause 7.2 requires carbon intensity to be applied to the energy consumed in each region. Sum energy by region:

E_server = Σ E_region_i

and either calculate emissions region by region, or use an energy-weighted average intensity, which gives the same result:

I_server = Σ (E_region_i × I_region_i) / Σ E_region_i
O_server = Σ (E_region_i × I_region_i)

Worked: 800 kWh in a region at 200 gCO2eq/kWh and 400 kWh in a region at 350 gCO2eq/kWh → 160 kg + 140 kg = 300 kg CO2eq; the weighted average intensity is 300,000 g ÷ 1,200 kWh = 250 gCO2eq/kWh. (illustrative placeholder, not a real measurement)

Serverless and edge functions​

  • Serverless function execution belongs in O_server.
  • Cold-start overhead is included: it is real resource consumption.
  • Edge function execution at CDN nodes belongs in O_server.
  • Clause 7.2 requires the allocation method for shared infrastructure to be disclosed. Serverless platforms rarely expose energy directly, so explain how you estimated it (for example from billed execution time and memory) and which coefficients you used.

Third-party server-side services​

Server-side APIs you call, such as a payment processor, are counted in O_server with a separate third-party sub-total (Clauses 5.4 and 7.6). See Third-Party Attribution for estimation methods.

Please submit any comments you have here.