Efficiency problems with MulticoreParam

multicore_efficiency.R

library(scRNAseq)
hESCs <- LaMannoBrainData('human-es')

# Trivial function to get something working on two matrices.
FUN <- function(x, y) { list(colSums(x), colSums(y)) }

library(BiocParallel)
BPPARAM1 <- SerialParam()
BPPARAM5 <- MulticoreParam(5)

# This function does nothing but split a matrix into chunks according
# to the number of cores it will eventually be parallelized over.
M <- SingleR:::.prep_for_parallel(assay(hESCs), BPPARAM=BPPARAM5)

bpstart(BPPARAM1)
system.time({
bplapply(M, FUN=FUN, y=assay(hESCs)) # 0.2 seconds
})
bpstop(BPPARAM1)

bpstart(BPPARAM5)
system.time({
bplapply(M, FUN=FUN, y=assay(hESCs), BPPARAM=BPPARAM5) # 2.6 seconds
})
bpstop(BPPARAM5)