Closed da-ekchajzer closed 2 years ago
https://ourworldindata.org/grapher/carbon-intensity-electricity from https://www.bp.com/en/global/corporate/energy-economics/statistical-review-of-world-energy.html (BP) TODO : Validate data
TODO : Find electricity primary energy factor by country (per year)
TODO : Find electricity abiotic depletion potential factor by country (per year)
If yearly_electrical_consumption is given by the user, use this value.
If not, yearly_electrical_consumption is measured as follows :
yearly_electrical_consumption = ((max_power*(1-idle_time_ratio)*medium_workload) + (idle_time_ratio * idle_power))*365*24
From Dell R640 LCA
_DEFAULT_IDLE_RATIO = 2.4/24 = 10%
_DEFAULT_IDLE_PERCENTAGE_MAX_POWER = 201/510 = 40%
_DEFAULT_MEDIUM_WORKLOAD = (100*3.6 + 50*13.2 + 10*4.8)/(24-2.4) = 50%
_DEFAULT_MAX_POWER = 510
if we apply the formula to the Dell R740:
yearly_electrical_consumption_dellR740
= ((_DEFAULT_MAX_POWER *(1-_DEFAULT_IDLE_RATIO )*medium_workload) +
(_DEFAULT_IDLE_RATIO * (_DEFAULT_IDLE_PERCENTAGE_MAX_POWER *_DEFAULT_MAX_POWER))
)*365*24
= 2189 kwh/year < 3081 kwh/year (Dell data)
I'm wondering what would be the most common situation from an API user POV when not having the electrical consumption (that would be most of the cases):
workload_ratio
attribute?) and the idle_time_ratio
if any + a number of hours/days it has run (it can be 365 days for on-prem always-on resources or something different for cloud resources)I think the ideal for us would be to compute for each server configuration the P (power consumption) values we see in the Dell example and then simply calculate the consumption based on the parameters or fall back to something we can define (365 days running an average load).
For cloud we will do an other process with a bottom-up approach per component #29 but the same problems occurs with the load
For server something like that :
LOAD | high (100%) | medium (50%) | low (10%) | idle | off |
---|---|---|---|---|---|
Power (W) | X | X | X | X | 0 |
1 - I have the different power per load -> Use it 2 - I only have the max power -> Can medium, low and idle be retrieve has function of max power ? 3 - I have nothing -> Use Dell R740 power per load
LOAD | high (100%) | medium (50%) | low (10%) | idle | off |
---|---|---|---|---|---|
Time_ratio (% of day per year or hour per day) | X | X | X | X | X |
1 - I have the different time ratio per load -> Use it 2 - I only have the medium workload -> what can we do ? 3 - I have nothing -> Use Dell R740 power per load
yearly_electrical_consumption = [power(high) * time_ratio(high) + power(medium) * time_ratio(medium) + power(low) * time_ratio(low) + power(idle) * time_ratio(idle) + power(off) * time_ratio(off)]*24*365
I think it is ok to use Dell R740 for now because in the future you will be able to select an archetype (#1). All the missing datas will be set with the value of this archetype and not the value of the default configuration.
Thanks,
For: "2 - I only have the max power -> Can medium, low and idle be retrieved as function of max power ?" we can define some heuristics based on existing data (to be discussed), maybe @bpetit would have some interesting thoughts on this from Scaphandre and we can reuse AWS bare-metal power profiles data on top of Dell's.
For: "3 - I have nothing -> Use Dell R740 power per load" Ok to use Dell as a first step and then the closest archetype profile available, to be discussed in (#1) but we can easily generate profiles like for example: 1 low TDP (<100W) CPU + 128 Gb memory etc.
Following today's meeting
We reach the same points (2 & 3)
We have identified the same problems with the time and affectation issues : When the user request the impacts for a specific time window how to affect the scope3 (linearly, all the impact, …)
Temporary decisions :
Power per load should be set as a percentage of max_power
Epic
Usage
US
As an external user I can retrieve the scope2 impact of devices With device usage and characteristics
Usage object
usage attributes
Yearly electrical consumption
yearly_electrical_consumption is given by the user or smart complete depending on the type of component or device (see below)
GWP
gwp = yearly_electrical_consumption*life_duration*carbon_intensity_factor
PE
pe = (yearly_electrical_consumption*life_duration) * primary_energy_factor
ADP
adp = (yearly_electrical_consumption*life_duration) * abiotic_depletion_potential_factor