Fitting lm on fake data using 2 nodes

doParallel_MultiNodes.r

# Parallelize linear regressions using 4 nodes

# Open an interactive job
# with the following command line
# salloc --time=1:0:0 --ntasks=100

#0. Packages loading
library(doParallel)

#1. Init the cluster of SOCKETS
# Definition: A network socket is an endpoint of an inter-process communication across a computer network.

# Get number of Cores by nodes based on SLURM ENV
nCores <- as.numeric(Sys.getenv("SLURM_CPUS_ON_NODE"))

# Get Nodes addresses
nodelist <- Sys.getenv("SLURM_NODELIST") 
nodelist <- paste0("cdr",unlist(stringi::stri_extract_all_regex(nodelist, "[0-9]+")))

# Build the cluster
cl <- makeCluster(rep(nodelist,nCores),type='SOCK')
registerDoParallel(cl)

#2. Generate fake tree dataset with 100 observations (volume) for 1.000 species and fit linear regression for each species
tree.df <- data.frame(species = rep(c(1:1000000), each = 100 ), girth = runif (100000000,7,40 ))
tree.df$volume <- tree.df$species / 10 + 5 * tree.df$girth + rnorm ( 10000000 , 0 , 3 )
# Extract species IDs to iterate over
species <- unique(tree.df$species)

#3. Run foreach loop and store results in fits object
fits <- foreach(i = species, .combine = rbind) %dopar% {
  sp <- subset (tree.df, subset = species == i)
  fit <- lm (volume ~ girth, data = sp)
  return ( c(i, fit $ coefficients))
}

#4. Stopping the cluster
stopCluster(cl)

Nice job !
It will doubtlessly be helpful for many R users !