Result of the bench

[DamienCassou]

Thank you very much for your tool. From the result, I guess zlib1 makes sense as it compresses compared to zlib0 and is not that bad compared to more aggressive compression schemes (which take much longer). What do you think?

[dragetd]

Uh, reading the code after learning python I am tempted to delete this repository and never show it anyone again. xD I should really rewrite the tests!

Anyways, concerning your results: Yes, zlib,0 is 'no compression' and sort of pointless.
In your case, zlib,1 would be the best time/rate trade off. lzma is extremly slow and only an option if space/bandwith is really important. Generally zlib,6 is a sane default for most users.

But since Borg 1.1 (which will be release very soon), Borg supports auto compression. It tries to compress the first few kilobytes with lz4 (which is very fast) to see if the file is compressible at all. If so, it uses a good compression. If not, it will use the fast lz4.

This should improve the time massively while keeping most if the compression benefits. I'll see if I can incorporate this into the benchmark when I ever get to cleaning up the code.