Benchmark

Michel Lang

2017-06-13

This small benchmark compares the performance of the base64 encoding/decoding in package base64url with the implementations in the packages base64enc and openssl.

Encoding of a single string

library(base64url)
library(base64enc)
library(openssl)
library(microbenchmark)

x = "plain text"
microbenchmark(
  base64url = base64_urlencode(x),
  base64enc = base64encode(charToRaw(x)),
  openssl = base64_encode(x)
)
## Unit: nanoseconds
##       expr   min      lq      mean  median      uq      max neval cld
##  base64url   611   680.0   1536.88   856.0  1000.5    39316   100   a
##  base64enc  1389  1661.5 174380.10  2209.5  2531.5 16820882   100   a
##    openssl 10218 10706.0  12939.33 11144.0 11711.0    98264   100   a

Decoding of a single string

x = "N0JBLlRaUTp1bi5KOW4xWStNWEJoLHRQaDZ3"
microbenchmark(
  base64url = base64_urldecode(x),
  base64enc = rawToChar(base64decode(x)),
  openssl = rawToChar(base64_decode(x))
)
## Unit: nanoseconds
##       expr   min      lq     mean  median      uq    max neval cld
##  base64url   707   862.0  1263.68  1161.5  1250.5  17633   100 a  
##  base64enc  4805  5141.0  6850.86  5753.5  6039.0 109692   100  b 
##    openssl 19169 19632.5 21585.91 19931.5 20854.0 104662   100   c

Encoding and decoding of character vectors

Here, the task has changed from encoding/decoding a single string to processing multiple strings stored inside a character vector. First, we create a small utility function which returns n random strings with a random number of characters (between 1 and 32) each.

rand = function(n, min = 1, max = 32) {
  chars = c(letters, LETTERS, as.character(0:9), c(".", ":", ",", "+", "-", "*", "/"))
  replicate(n, paste0(sample(chars, sample(min:max, 1), replace = TRUE), collapse = ""))
}
set.seed(1)
rand(10)
##  [1] "zN.n9+TRe"                     "mVA1IX/"                      
##  [3] "1,oSisAaA8xHP"                 "m5U2hXC4S2MK2bGY"             
##  [5] "G7EqegvJTC.uFwSrH0f8x5x"       "G97A1-DXBw0"                  
##  [7] "XiqjqeS"                       "13FC3PTys/RoiG:P*YyDkaXhES/IH"
##  [9] "0FJopP"                        "fcS,PMK*JVPqrYFmZh7"

Only base64url is vectorized for string input, the alternative implementations need wrappers to process character vectors:

base64enc_encode = function(x) {
  vapply(x, function(x) base64encode(charToRaw(x)), NA_character_, USE.NAMES = FALSE)
}

openssl_encode = function(x) {
  vapply(x, function(x) base64_encode(x), NA_character_, USE.NAMES = FALSE)
}

base64enc_decode = function(x) {
  vapply(x, function(x) rawToChar(base64decode(x)), NA_character_, USE.NAMES = FALSE)
}

openssl_decode = function(x) {
  vapply(x, function(x) rawToChar(base64_decode(x)), NA_character_, USE.NAMES = FALSE)
}

The following benchmark measures the runtime to encode 1000 random strings and then decode them again:

set.seed(1)
x = rand(1000)
microbenchmark(
  base64url = base64_urldecode(base64_urlencode(x)),
  base64enc = base64enc_decode(base64enc_encode(x)),
  openssl = openssl_decode(openssl_encode(x))
)
## Unit: microseconds
##       expr       min        lq       mean    median        uq        max
##  base64url   203.969   271.417   371.5827   349.551   447.294    912.724
##  base64enc  6315.815  6873.980  7805.2762  7560.447  8360.947  13229.642
##    openssl 25749.224 27296.567 29627.6411 28116.323 29651.793 111755.171
##  neval cld
##    100 a  
##    100  b 
##    100   c