Chorba: A novel CRC32 implementation (2024)
arxiv.org59 points by fnands 3 days ago
59 points by fnands 3 days ago
This repo's readme gives a great overview of the previous-best approaches (as of ~6 years ago): https://github.com/komrad36/CRC
Warning: the README is 134KB... It's good but, like me, you might be here a while.
In Spain, "chorba" is very informal slang for "gal" [0]. Not vulgar, just very informal vernacular.
Chorba is also soup in Eastern European languages like Bulgarian and Romanian.
And also in Turkey. It (the word, if not the stew itself) arrived to the Balkans by way of the Ottomans. (And having just now clicked through to the link, it seems to have arrived to Turkey by way of the Persians).
and is a traditional Tunisian soup
/Edit: actually in all North Africa https://en.wikipedia.org/wiki/Chorba
All this talk of soup making me wonder if these are Arabic/ME derivatives.
Then again there are like 10 different ways to refer to soup in the various dialects.
The wikipedia article traces it to Persian, which formed it as a compound of words from different East Iranian languages. So you are on the money with Middle Eastern. From there it spread to the Balkan via Ottoman Turkish, and also from Persian to dialectal Arabic, which would explain the occurrences in Northern Africa, and maybe even Spain
In Europe this guy stands for vulgar and racist neofascism
If you have a problem with my comments, there is a link in my profile to a convenient way to hide comments from people you don't like. You are welcome!
> Dedication
> This implementation is named after the Serbian singer Bora Đorđević (also known as Bora Čorba) who was born in 1952 and died in 2024. His birth year matches the number of the GZIP standard RFC 1952 that describes a common CRC32 implementation, and the original proof of concept for this method used the polynomial x21 +x15 + x14 + x11 + x10 + x7 + x3 which is x1952×8 mod G(x).
That is indeed dedication.
What are the units on the vertical axes for figures 1 and 2? I might have guessed seconds per TiB but the braiding line doesn't seem to match what's in figure 3.
I forget, these were the outputs of my test scripts and I wasn't exactly fastidious here, at least the axes all start from zero :)
IIRC they're off by maybe a factor of 10 or 100, the test scripts just generate a bunch of (seeded) random data and then execute the CRC of X bits Y times and that's where the number comes from, it's consistent across the different tests even if the units are wrong
Anyone can replicate the results? In any case, works like this give me moments of epiphany when I start to believe the humanity is not totally lost.
zlib-ng implements it so I assume it is practical at the very least
the zlib-ng guys are super welcoming and helpful, working with them was a very pleasant experience
News to me, but a guy named Sam Russell came up with a new software only CRC32 algorithm that is competitive with hardware accelerated implementations. It's a surprisingly elegant solution.