Gzip 1.11 Released With "Orders of Magnitude Faster" Performance On IBM Z

[phoronix]

Phoronix: Gzip 1.11 Released With "Orders of Magnitude Faster" Performance On IBM Z

The last release of Gzip was v1.10 back at the end of 2018 while now it's finally been succeeded today by Gzip 1.11...

https://www.phoronix.com/scan.php?page=news_item&px=Gzip-1.11-Released

[ddriver]

Better than zstd?

[kieffer]

I have been testing their hardware compression engine on IBM power9 and basically their hardware gzip is at the speed of LZ4 !
See slide #12 of https://www.hdfgroup.org/wp-content/...Power9HDF5.pdf

[linner]

But gzip is not a compression algorithm, it's a file format. Doesn't gzip just use zlib?

[KesZerda]

But gzip is not a compression algorithm, it's a file format. Doesn't gzip just use zlib?

No, gzip is a completely separate implementation of the DEFLATE algorithm, that is ever so subtly different than the default settings for zlib -- gzip defaults to a some extra header and footer info that zlib's implementation doesn't provide.

[microcode]

I think it would be cool to see hardware zstd, because the software decompression speed is excellent, and the ratios are often like LZMA (depends on the data). Ditto hardware ChaCha: yes, it's fast in software, but imagine how fast and power efficient it could be in hardware when compared with hardware AES. I wish the push was more toward hardware ChaCha than hardware AES, because it would assure better performance across the board, including places where hardware AES is not available on the far side of a channel.