parallel

parallel

Base R parallel computing support.

Detect Cores

library(parallel)

# Number of cores
detectCores()
detectCores(logical = FALSE)  # Physical cores only

mclapply (Unix/Mac only)

# Parallel lapply
result <- mclapply(1:100, function(x) x^2, mc.cores = 4)

# With more options
result <- mclapply(
  X = data_list,
  FUN = process_function,
  mc.cores = detectCores() - 1,
  mc.preschedule = TRUE,
  mc.set.seed = TRUE
)

parLapply (Cross-platform)

# Create cluster
cl <- makeCluster(4)

# Parallel lapply
result <- parLapply(cl, 1:100, function(x) x^2)

# Stop cluster
stopCluster(cl)

Cluster Types

# PSOCK (default, cross-platform)
cl <- makeCluster(4, type = "PSOCK")

# FORK (Unix/Mac only, shares memory)
cl <- makeCluster(4, type = "FORK")

# MPI cluster
cl <- makeCluster(4, type = "MPI")

Export Variables

cl <- makeCluster(4)

# Export variables to workers
clusterExport(cl, c("my_data", "my_function"))

# Export from specific environment
clusterExport(cl, "var", envir = my_env)

# Evaluate expression on workers
clusterEvalQ(cl, library(dplyr))

result <- parLapply(cl, data_list, my_function)
stopCluster(cl)

Parallel Apply Functions

cl <- makeCluster(4)

# parLapply - parallel lapply
parLapply(cl, X, FUN)

# parSapply - parallel sapply
parSapply(cl, X, FUN)

# parApply - parallel apply for matrices
parApply(cl, matrix, MARGIN, FUN)

# parRapply - parallel row apply
parRapply(cl, matrix, FUN)

# parCapply - parallel column apply
parCapply(cl, matrix, FUN)

stopCluster(cl)

Load Balancing

cl <- makeCluster(4)

# Static scheduling (default)
parLapply(cl, X, FUN)

# Dynamic load balancing
parLapplyLB(cl, X, FUN)
parSapplyLB(cl, X, FUN)

stopCluster(cl)

Random Number Generation

cl <- makeCluster(4)

# Set up parallel RNG
clusterSetRNGStream(cl, iseed = 123)

# Now random numbers are reproducible
result <- parLapply(cl, 1:10, function(x) rnorm(1))

stopCluster(cl)