peakRAM for loops

[danxfreeman]

Thanks for developing this @tpq, I've been using it to benchmark a package I'm developing against a current one. One question though - does peakRAM work for loops? I'm trying to understand why these two calls give different outputs.

mem <- peakRAM({
  for(i in 1:10) {
    mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 10008235

mem <- peakRAM({
  for(i in 1:5) {
    mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 5006635, shouldn't this be the same?

[tpq]

Thanks Dan! I'm glad you've found the package helpful. I made it exactly for that purpose!

This is a very interesting discovery, and I'm afraid that I don't have a very good answer. I am the first to admit that peakRAM is a little bit hacky. It basically works in 4 steps: (1) Run the garbage collector to reset, (2) Call the function, (3) Run the garbage collector again, (4) Take the difference of Step 3 and Step 1.

It compares the before & after RAM usage recorded by the garbage collector as a proxy for the actual RAM used. I have read that this can be misleading at times. I am unsure whether the for-loop is one example.

Interestingly, when I run it locally I get slightly different results:

mem <- peakRAM({
  for(i in 1:2) {
    x <- mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 15.3

mem <- peakRAM({
  for(i in 1:5) {
    x <- mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 38.2

mem <- peakRAM({
  for(i in 1:10) {
    x <- mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 53.4

mem <- peakRAM({
  for(i in 1:20) {
    x <- mean(rnorm(1e6))
  }
})
mem$Peak_RAM_Used_MiB # 53.4

(FWIW I am using the latest version on GitHub with R 3.6.3)