RAM coefficient value - implementation vs. HotCarbon paper

[davidkopp]

At the moment, the coefficient value 0.284 W/GB is used in the implementation.

Source code:

boaviztapi/boaviztapi/model/consumption_profile/consumption_profile.py

Line 30 in 4af9f11

ram_electrical_factor_per_go = 0.284

Docs: https://doc.api.boavizta.org/Explanations/components/ram/#determining-the-parameters

In the paper that was recently published during the HotCarbon conference a different value is mentioned:

For a given model, the TechPowerUp SSD specs database [35] can be used to estimate the idle and active (averaged) power consumption values. For RAM banks, we use the averaged values from [33]: 0.19 W /GB and 0.54 W /GB in idle and active states, respectively. By default, RAM banks are constantly considered in an active state.

Source: Simon, T. et al. (2024) ‘BoaviztAPI: a bottom-up model to assess the environmental impacts of cloud services’, in. HotCarbon 2024. Available at: https://hotcarbon.org/assets/2024/pdf/hotcarbon24-final74.pdf

This seems to me as an inconsistency. Which value is more accurate?

[da-ekchajzer]

@tibosmn @samuelrince do you remember ? Should we update the value ?

[tibosmn]

If I recall correctly, we decided to use a value sourced from the literature rather than one derived from the modeling process, which primarily show a static consumption anyway.
I'd update the value but let the final call to @samuelrince ;)

[samuelrince]

Indeed, the current value was an average computed from @github-benjamin-davy AWS dataset. But, we chose a different source in the paper. I think we should update the value in the API.

To do:

• Change RAM consumption profile to only support idle and active states
• Allow user to change, idle and active power consumption in addition to an utilization ratio
• Compute that utilization ratio for archetypes depending on the instance type
• Have a default value for the utilization set to 50%, maybe?