Replies: 3 comments 2 replies
-
Cloudflare (cloudflare/zlib@959b4ea)
My comments:The Cloudflare fork clearly shows that it is possible to optimize parts of the code further, kudos to them for that. 🚀 IMO, the big disadvantage with the Cloudflare fork is in the quality of the code, build-system and (lack of public) testing:
|
Beta Was this translation helpful? Give feedback.
-
@Dead2 - Any advice on benchmarking difference Zstandard vs Zlib-ng ? |
Beta Was this translation helpful? Give feedback.
-
A comparison against https://github.com/google/zopfli would also be interesting |
Beta Was this translation helpful? Give feedback.
-
Benchmark method details
Benchmarked using deflatebench (zlib-ng/deflatebench@4e65b35).
Compression data used is Silesia-small.tar (see README.md from https://github.com/zlib-ng/deflatebench).
Benchmarks run on a Scientific Linux 7 system with a custom 5.10.16 kernel.
Compiled using stock compiler GCC 4.8.5 (gcc-4.8.5-44.el7.x86_64)
CPU: i9-9900K, 32GB RAM (2x Crucial CT16G4DFD8266), Motherboard: Supermicro C9Z390-CGW
System has several tweaks to avoid noise, with benchmarking processes assigned to 2 dedicated cpu-cores that have been isolated+nohz_full by kernel, as well as the system running the bare minimum of background processes. HyperThreading partner cores offlined. CPU turbo and speedstep disabled. Input and output files reside in tmpfs. Etc..
My comments to these results:
zlib-ng (aeffa9b) with zlib-ng's minigzip
zlib (cacf7f1) with zlib's minigzip
zlib (cacf7f1) with zlib-ng's minigzip (aeffa9b)
gzip (gzip-1.5-10.el7.x86_64)
Beta Was this translation helpful? Give feedback.
All reactions