Allocation methodology

Allocation Methodology #

One key element that differentiates levv cloud is the ability to provide you with impact metrics of your applications so that you can optimize your applications with regard to impact.

This data is based on our energy measurements and an allocation model used to match a proportion of our total energy usage to the usergy usage of your application.

The data we provide you depends heavily on our allocation model, which might evolve, and we present it here below.

Methodology #

The Green Software Foundation has developed a methodology, the SCI framework, to calculate the rate of carbon emissions for software systems. The methodology is based on a formula that takes into consideration different factors that contribute to emissions. It can be summarized as

SCI = (O + M) per R

Where

  • O is equivalent to the operational emissions.
  • M is equivalent to the embodied hardware emissions.
  • R is a rate, i.e., the defined boundaries for which this observations are valid.

For levv, R is a time boundary, given in hours of operation, as this is recommended for cloud computing services. In our calculations, SCI results are in grams of CO2 equivalent (gCO2eq) per hour.

O: Operational Emissions

The O in the equation can be further discerned into

O = E x I

Where

E: Energy

We use Kubernetes-based Efficient Power Level Exporter, or Kepler, to obtain the energy consumed by applications running in the cluster.

I: Carbon Intensity

As the data center is located in Brussels, we use the average power intensity for Belgium in 2023, obtained from Ember and made available by the Green Web Foundation. The value is currently set at 137.68 gCO2eq per kWh.

Grid intensity plays a huge factor on this calculation, for example, some nordic countries have currently a value of 7 gCO2eq per kWh, only 5% of current value for Belgium.

Note: In the future, this value will be adapted to the one given by the energy provider for the data center.

Power Usage Efectiveness: PUE #

Cloud computing operations rely on data centers to provide services. Power usage effectiveness (PUE) is a ratio that describes how efficiently a data center uses energy, specifically by comparing how much of the energy consumed by the data center is used on computing in contrast to supporting the facilities. Anything that isn’t considered a computing device in a data center (e.g. lighting, cooling, etc.) falls into the category of facility energy consumption.

PUE = Total Facility Energy / IT Equipment Energy = 1 + Non IT facility Energy / IT Equipment Energy

An ideal PUE is 1.0.

PUE affects the operational emissions calculation.

M: Embodied Hardware Emissions

Embodied carbon is the amount of carbon emitted during the creation and disposal of a hardware device. Within the SCI, a fraction of the total embodied emissions of the hardware is allocated to the software.

To calculate the share of embodied emissions in gCO2eq, we can use

M = TE x TS x RS

Where

  • TE is the total emissions for manufacturing the hardware components.
  • TS is the time share, a factor that correlates used time and total life span of the hardware.
  • RS is the resource share, a factor indicating how many resources, of the total, the software needs to run.

This can be further expanded into

M = TE x TiR / EL x RR / ToR

Where

  • TiR is time reserved, or the length of time reserved for the software use.
  • EL is expected lifespan of the hardware.
  • RR is resources reserved, the number of resources reserved for the use of the software, for example: CPU units, gigabytes of RAM.
  • ToR is total resources of the hardware, for example: total CPU units, total gigabytes of RAM.

TE: Manufacturing emissions

We used a an web tool provided by Boavizta to approximate the manufacturing emissions corresponding to our hardware.

SSDs account for 45% of the total embodied carbon and is by far the largest contributor in our hardware.

RS: Resource Share

We base this on the size of your deployment and rely on CPU only as it is the main contributor to power consumption (compared to RAM and storage).