I just noticed it is as "lzip" but also make an entry for 7zip using lzma on ultra
Archive Compression Schemes
Levi White
Hunter Lopez
zstd is the mediocre in all dimensions. It will sometimes be best depending on the situation. In particular, suppose you want to send data over a wire as fast as possible. The time taken is d + s/b + c (d:decompression time, c:compression time, b: byterate, s:compressed size). Graphing that:
b(B/s) 1.00E+03 1.00E+06 1.00E+09paq 1.48E+04 8.76E+03 8.75E+037z1 1.16E+04 2.79E+01 1.63E+01bz 1.22E+04 1.72E+01 5.02E+00lz 1.26E+04 3.17E+01 1.91E+01plz 1.26E+04 3.22E+01 1.96E+017z2 1.27E+04 2.18E+01 9.09E+00xz 1.27E+04 2.86E+01 1.59E+01zstd 1.39E+04 2.30E+01 9.12E+00gz 1.75E+04 2.09E+01 3.50E+00none 3.54E+04 3.54E+01 3.54E-02
at 1k/s 7z -m0=ppmd is best. at 1M/s bz wins. at 1G/s, no compression is best, second best is gz. On this graph, I see nowhere that zstd is best, but lets add another one assuming that compression happens once, but downloading +decompression happens n times (ie d+(s/b+c)*n) (also standardize on b=1e6):
n 1.00E+01 1.00E+03 1.00E+05 1.00E+07paq 8.81E+03 1.48E+04 6.15E+05 6.06E+077z1 2.08E+02 2.01E+04 2.01E+06 2.01E+08bz 1.40E+02 1.37E+04 1.37E+06 1.37E+08lz 1.54E+02 1.36E+04 1.36E+06 1.36E+08plz 1.56E+02 1.37E+04 1.37E+06 1.37E+087z2 1.43E+02 1.35E+04 1.35E+06 1.35E+08xz 1.51E+02 1.36E+04 1.36E+06 1.36E+08zstd 1.50E+02 1.41E+04 1.41E+06 1.41E+08gz 1.81E+02 1.78E+04 1.77E+06 1.77E+08none 3.54E+02 3.54E+04 3.54E+06 3.54E+08
For ten downloads bz is best. at 1000 they're all identical lol, but 7z2 is best. at 100,000 paq starts throwing it's weight, the rest are all identical again, same for 10,000,000. The lesson basically is that compression time matters very little, of primary importance (for downloads) is ratio and decompression time.
>done in excel
For archival, literally the only thing that matters is ratio (and reliability). You can run it over night if it's too slow, along with your indexer and defragger.
Ryder Peterson
...
Asher Perez
More charts:
* text: a ~4.4M syslog* photo:> du -h -- "Honda CR-X front-left.ppm"18M Honda CR-X front-left.ppm> file -b -- "Honda CR-X front-left.ppm"netpbm image data, size = 3008 x 2000, rawbits, pixmap* flat image:> du -h -- "Satanichia smug.ppm"4.1M Satanichia smug.ppm> file -b -- "Satanichia smug.ppm"Netpbm image data, size = 1400 x 1000, rawbits, pixmap* audio: > du -h -- "10. Song of Sirens.wav"27M 10. Song of Sirens.wav> file -b -- "10. Song of Sirens.wav"RIFF (little-endian) data, WAVE audio, Microsoft PCM, 16 bit, stereo 44100 Hz
Some remarks, ffv1 looks very good as a quite slow PNG replacement (considering it's probably not as optimized) and jpeg2000 is pretty impressive for a "forgotten" codec.
Charles King
Looks like the photo one didn't upload.
Owen Kelly
Fyi, you can upload your scripts, results, and test suites ITT as ".lzma.pdf".
Your images really mean nothing if we can't replicate your results, esp. on different hardware.
Jacob Brooks
p7zip has not been updated in 2 years, while 7zip is in active development and its last release was 6 months ago.
As you can see at
7-zip.org
there were quire a few bugfixes after 17.00beta
Caleb Clark
How the fuck is jpeg2000 forgotten when the entire movie industry uses when releasing to theaters? If you want to include obscure compression formats, switch FLAC with TAK as it compresses better, but practically nobody uses it.
Carson Martinez
So does that mean it doesn't work any more?
I'd better alert all my 7z archives that they're old and broken.
Eli Stewart
m8, I can't even post with Tor on this shitty imageboard (I dided yesterday because the onion service was down). Anyway, I could at least post the images if someone was interested but nobody is, that's why I didn't bother.