A quick comparison between fst and rds IO characteristics

fst_rds_benchmark.R

library(fst)

n <- 1000000
d1 <- as.data.frame(c(replicate(50, sample(letters[1:10], n, replace = TRUE), simplify = FALSE), 
                      replicate(50, runif(n), simplify = FALSE)))
tmpf <- tempfile()

rw_time <- function(compression, file, write.fun, read.fun) {
  data.frame(compression = compression,
             writeTime = system.time(write.fun(d1, file, compress = compression))[[3]],
             fileSize = paste(round(file.size(file)/1024^2, 2), "Mb"),
             readTime = system.time(read.fun(file))[[3]])  
}

## fst
do.call(
  rbind,
  lapply(seq(50, 100, 10), 
        rw_time, 
        write.fun = write_fst, 
        read.fun = read_fst,
        file = tmpf))

# compression writeTime  fileSize readTime
# 1          50     0.248 430.42 Mb    0.119
# 2          60     0.319  397.7 Mb    0.176
# 3          70     0.614  364.1 Mb    0.147
# 4          80     1.217 329.85 Mb    0.138
# 5          90     3.980 294.82 Mb    0.149
# 6         100     5.985 260.24 Mb    0.226

## rds
do.call(
  rbind,
  lapply(c(FALSE, TRUE), 
         rw_time, 
         write.fun = saveRDS, 
         read.fun = readRDS,
         file = tmpf))
# compression writeTime  fileSize readTime
# 1       FALSE     0.551 572.23 Mb     0.52
# 2        TRUE    48.262 284.44 Mb     2.28