Compression Benchmarks: brotli, gzip, xz, bz2

[This article was first published on OpenCPU, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

opencpu logo Brotli is a new compression algorithm optimized for the web, in particular small text documents. Brotli decompression is at least as fast as for gzip while significantly improving the compression ratio. The price we pay is that compression is much slower than gzip. Brotli is therefore most effective for serving static content such as fonts and html pages. The brotli package is now on CRAN and supports both compression and decompression of the brotli format. Let’s benchmark the available compression formats in R using a some example text data from the COPYING file.
library(brotli)
library(ggplot2)

# Example data
myfile <- file.path(R.home(), "COPYING")
x <- readBin(myfile, raw(), file.info(myfile)$size)

# The usual suspects
y1 <- memCompress(x, "gzip")
y2 <- memCompress(x, "bzip2")
y3 <- memCompress(x, "xz")
y4 <- brotli_compress(x)
Confirm that all algorithms are indeed lossless:
stopifnot(identical(x, memDecompress(y1, "gzip")))
stopifnot(identical(x, memDecompress(y2, "bzip2")))
stopifnot(identical(x, memDecompress(y3, "xz")))
stopifnot(identical(x, brotli_decompress(y4)))

Compression ratio

If we compare compression ratios, we can see Brotli significantly outperformes the competition for this example.
# Combine data
alldata <- data.frame (
  algo = c("gzip", "bzip2", "xz (lzma2)", "brotli"),
  ratio = c(length(y1), length(y2), length(y3), length(y4)) / length(x)
)

ggplot(alldata, aes(x = algo, fill = algo, y = ratio)) + 
  geom_bar(color = "white", stat = "identity") +
  xlab("") + ylab("Compressed ratio (less is better)")
brotli compression ratio

Decompression speed

Perhaps the most important performance dimension for internet formats is decompression speed. Clients should be able to decompress quickly, even with limited resources such as on browsers and mobile devices.
library(microbenchmark)
bm <- microbenchmark(
  memDecompress(y1, "gzip"),
  memDecompress(y2, "bzip2"),
  memDecompress(y3, "xz"),
  brotli_decompress(y4),
  times = 1000
)

alldata$decompression <- summary(bm)$median
ggplot(alldata, aes(x = algo, fill = algo, y = decompression)) + 
  geom_bar(color = "white", stat = "identity") +
  xlab("") + ylab("Decompression time (less is better)")
brotli decompression speed We see that brotli is very similar to gzip in decompression speed. We also see why bzip2 and xz have never replaced gzip as the standard compression method on the internet, even though they have better compression ratio: they are several times slower to decompress.

Compression speed

So far Brotli showed the best compression ratio, with decompression performance comparable to gzip. But there is no such thing as a free pastry in Switzerland. Here is the caveat: compressing data with brotli is complex and slow:
library(microbenchmark)
bm <- microbenchmark(
  memCompress(x, "gzip"),
  memCompress(x, "bzip2"),
  memCompress(x, "xz"),
  brotli_compress(x),
  times = 20
)

alldata$compression <- summary(bm)$median
ggplot(alldata, aes(x = algo, fill = algo, y = compression)) + 
  geom_bar(color = "white", stat = "identity") +
  xlab("") + ylab("Compression time (less is better)")
brotli compression speed Hence we can conclude that Brotli is mostly nice for clients, with decompression performance comparable to gzip while significantly improving the compression ratio. These are powerful properties for serving static content such as fonts and html pages. However compression performance, at least for the current implementation, is considerably slower than gzip, which makes Brotli unsuitable for on-the-fly compression in http servers or other data streams.

To leave a comment for the author, please follow the link and comment on their blog: OpenCPU.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)