Dev1 Galaxy Forum / Refractasnapshot – Comparison of different Compression Algorithms

Re: Refractasnapshot – Comparison of different Compression Algorithms

dummy@example.com (Xaver) — Mon, 13 Jul 2026 15:34:30 +0000

Now I have integrated these insights in /etc/refractasnaoshot.conf
The compression sections of my refractasnaoshot.conf now looks like this:

-----------------------------
[...]

# Uncomment one of the lines below to use xz or zstd compression for smaller iso.

# small and slow # relative size of the iso-image
#mksq_opt="-comp xz" # (-4.1% to zstd -19 | -15,4% to zstd -1)
#mksq_opt="-comp xz -b 262144" # (-2.0% to xz)
#mksq_opt="-comp xz -b 524288" # (-3.8% to xz)
#mksq_opt="-comp xz -b 1048576" # (-5.5% to xz)

# smaller and slower:
#mksq_opt="-comp xz -Xbcj x86" # (-1.5% to xz)
#mksq_opt="-comp xz -Xbcj x86 -b 524288" # (-5.4% to xz)
#mksq_opt="-comp xz -Xbcj x86 -b 1048576" # (-6.9% to xz)

# not as small but faster
#mksq_opt="-comp zstd -Xcompression-level 22" # (= zstd -19 | --ultra ?)
#mksq_opt="-comp zstd -Xcompression-level 19" # (-11.8% to zstd -1)
mksq_opt="-comp zstd -Xcompression-level 15" # (-11.1% to zstd -1)
#mksq_opt="-comp zstd -Xcompression-level 8" # (-7.0% to zstd -1)
#mksq_opt="-comp zstd -Xcompression-level 3" # (-3.1% to zstd -1)
#mksq_opt="-comp zstd -Xcompression-level 1" # (-17.2% to lz4)

#mksq_opt="-comp lz4" # (+20,8% to zstd -1)

[...]

Re: Refractasnapshot – Comparison of different Compression Algorithms

dummy@example.com (fsmithred) — Thu, 02 Jul 2026 16:35:45 +0000

Thanks for that info. I made a note to myself to try a few different settings when I next make some live-isos. Faster builds and faster boots would be nice.

Refractasnapshot – Comparison of different Compression Algorithms

dummy@example.com (Xaver) — Sun, 28 Jun 2026 17:29:11 +0000

When I was creating an iso-image of my new Refracta based system, I have noticed, that zstd -22 and zstd -19 deliver the same results. Obviously the zstd ultra mode does not work properly within Refractasnapshot. This inspired me to do some testing of different compression options.
Here are my results:

Total size of the system on qemu hd:
/ 4,09 GB + /home/ 289 MB = 4,37 GB

xz: 1572 MB (-4.1% to zstd -19 | -15,4% to zstd -1)
xz -b 262144: 1540 MB    (-2.0% to xz)
xz -b 524288: 1512 MB    (-3.8% to xz)
xz -b 1048576: 1486 MB    (-5.5% to xz)

xz -Xbcj x86 1548 MB (-1.5% to xz)

zstd -22: 1639 MB ('zstd --ultra -22' does not work)
zstd -19: 1639 MB (-11.8% to zstd -1)
zstd -15: 1652 MB (-11.1% to zstd -1)
zstd -8: 1727 MB (-7.0% to zstd -1)
zstd -3: 1800 MB (-3.1% to zstd -1)
zstd -1: 1858 MB (-17.2% to lz4)

lz4: 2245 MB (+20,8% to zstd -1)

Relative compression speed according to Allan Jude (2017):
[https://openzfs.org/w/images/b/b3/03-Op … in_ZFS.pdf]
zstd -19 = 1x | zstd --ultra -22 = 0.58X
zstd -15 = 2.3x | zstd -8 = 16.1x
zstd -3 = 59.1x | zstd -1 = 101.5x
lz4 = 172.3x

According to Northon Torga (2024) xz and zstd -19 have a similar compression speed:
[https://ntorga.com/gzip-bzip2-xz-zstd-7z-brotli-or-lz4/]
zstd -19 = 1x | xz = 0.96x

Decompression speed of xz is about 5 times slower than zstd:
xz = ~0,18x | zstd = 1x | lz4 = ~ 3.2x

Depending on hardware and kind of files the actual results may differ.