greenAlgoR

This package computes ecological footprint in R (based on green-algorithms, Lannelongue, Grealey, and Inouye (2021)). The aim is to facilitate the programmatic used of green-algorithms using the function ga_footprint(). Moreover, greenAlgoR made it simple to compute ecological footprint of {targets} pipelines using function ga_targets().

It is a beta-version. Please submit issues if you found problems or if you want to contribute.

Installation

greenAlgoR is not available on CRAN for the moment. You can install the stable development version from GitHub with:

if (!require("devtools", quietly = TRUE)) {
  install.packages("devtools")
}
devtools::install_github("adrientaudiere/greenAlgoR")

Scientific foundation

The algorithm from Lannelongue, Grealey, and Inouye (2021)

You can refer to green-algorithms and Lannelongue, Grealey, and Inouye (2021) to understand the algorithm. Here is the short description from green-algorithms

The carbon footprint is calculated by estimating the energy draw of the algorithm and the carbon intensity of producing this energy at a given location:

$$carbon footprint = energy needed * carbon intensity$$

Where the energy needed is:

$$runtime * (power draw for cores * usage + power draw for memory) * PUE * PSF$$

The power draw for the computing cores depends on the model and number of cores, while the memory power draw only depends on the size of memory available. The usage factor corrects for the real core usage (default is 1, i.e. full usage). The PUE (Power Usage Effectiveness) measures how much extra energy is needed to operate the data centre (cooling, lighting etc.).

The PSF (Pragmatic Scaling Factor) is used to take into account multiple identical runs (e.g. for testing or optimisation). The Carbon Intensity depends on the location and the technologies used to produce electricity.

Basic use

devtools::load_all()
library("greenAlgoR")

In classical workflow

Using explicit input

# Your model must match exactly a name in TDP_cpu_internal$model
fp_12h <- ga_footprint(runtime_h = 12, cpu_model = "Core i3-10300")
fp_12h$carbon_footprint_total_gCO2
#> [1] 829.0232
fp_12h$energy_needed_kWh
#> [1] 1.745312

ggplot(fp_12h$ref_value, aes(
  y = variable,
  x = as.numeric(value),
  fill = log10(prop_footprint)
)) +
  geom_col() +
  geom_col(
    data = data.frame(
      variable = "Total",
      value = fp_12h$carbon_footprint_total_gCO2
    ),
    fill = "grey30"
  ) +
  geom_col(
    data = data.frame(
      variable = "Cores",
      value = fp_12h$carbon_footprint_cores
    ),
    fill = "darkred"
  ) +
  geom_col(
    data = data.frame(
      variable = "Memory",
      value = fp_12h$carbon_footprint_memory
    ),
    fill = "orange"
  ) +
  scale_x_continuous(
    trans = "log1p",
    breaks = c(0, 10^c(1:max(log1p(as.numeric(fp_12h$ref_value$value)))))
  ) +
  geom_vline(
    xintercept = fp_12h$carbon_footprint_total_gCO2,
    col = "grey30",
    lwd = 1.2
  ) +
  geom_label(
    aes(label = round_conditionaly(prop_footprint)),
    fill = "grey90",
    position = position_stack(vjust = 1.1)
  ) +
  labs(
    title = "Carbon footprint of the analysis",
    subtitle = paste0("(", fp_12h$carbon_footprint_total_gCO2, " g CO2", ")"),
    caption = "Please cite Lannelongue et al. 2021 (10.1002/advs.202100707)"
  ) +
  xlab("Carbon footprint (g CO2) in log10") +
  ylab("Modality") +
  theme(legend.position = "none")

Based on the R session

fp_session <- ga_footprint(runtime_h = "session", add_storage_estimation = TRUE)
fp_session$carbon_footprint_total_gCO2
#>  user.self 
#> 0.02016005
fp_session$energy_needed_kWh
#>    user.self 
#> 4.244221e-05

Based on a targets pipeline

ga_targets()

Roadmap

• Automatically find cpu model using benchmarkme::get_cpu()
• Submit to CRAN?
• Allow to ovewrite carbone intensity (e.g. with values from https://app.electricitymaps.com/)

Lannelongue, Loïc, Jason Grealey, and Michael Inouye. 2021. “Green Algorithms: Quantifying the Carbon Footprint of Computation.” Advanced Science 8 (12): 2100707. https://doi.org/10.1002/advs.202100707.