Yonatan Koren

Thoughts and coding efforts

yonkornilov commit calendar

Switching to lzop for fast decompression2017-09-14

When I provision my instances, I need to download an archive of a translation engine and extract it. Some options that I've tried:

I was recommended to try lzop as the decompression speed is allegedly much faster.

I created an archive of the directory default:

tar c default | lzop - > default.tar.lzo

I compared the size to the .gz archive:

# stat default.gz | grep 'Size'
Size: 6648380497 Blocks: 12993623   IO Block: 131072 regular file
# stat default.tar.lzo | grep 'Size'
Size: 8810709804 Blocks: 17219473   IO Block: 131072 regular file

Clearly the .lzo archive is larger. But perhaps the decompression speed can make it worth it for my use case, so let's time it.

We also have to pipe it through tar in order to keep the directory structure, since it's not a single file:

# time lzop -d -c default.tar.lzo | tar xv
real    1m18.581s
user    0m50.349s
sys     0m33.539s
# time tar xvzf default.gz
real    2m23.931s
user    2m2.211s
sys     0m38.708s

Clearly it's significant. If you favour speed over compression ratio, you should definitely use lzop.

Apparently there's a way to make the decompression even faster. Let's try it out:

You need to add the --fast flag while compressing:

tar c default | lzop --fast - > default.tar.lzo

# time lzop -d -c default.tar.lzo | tar xv
real    1m21.701s
user    0m45.270s
sys     0m46.466s
# stat default.tar.lzo | grep 'Size' 
Size: 8855209044 Blocks: 17306449   IO Block: 131072 regular file

Surprisingly this run was even slower than the first lzop run. However the archive is slightly larger in size when you use the --fast flag. I recommend using the default, i.e. not using the flag.