zfs

zfs copied to clipboard

Here is some test runs:

https://billatnapier.medium.com/falkhash-exotic-probably-shitty-hash-makes-me-smile-df1bbe2469dd

Github

https://github.com/gamozolabs/falkhash

But meow hash is according author 10% faster than falkhash, but its written in cpp:

https://github.com/cmuratori/meow_hash

And here according research paper faster than meow and falkhash:

https://github.com/jbapple/HalftimeHash

Pdf: https://arxiv.org/pdf/2104.08865

Thanks and

Best regards

So, while it undoubtedly is probably a stronger hash, I do wonder if those AES rounds with AES-NI are comparably as fast as the SSE/AVX accelerated fletcher 4 checksums, which do not have to rebase (unlike say the adler32 checksum).

It seems that meow and SHAMATA-256 are also this type of AES-NI hashing. They both got some Cryptanalysis, which seem not okay for our use case:

• Cryptanalysis for MEOW
• Cryptanalysis for SHAMATA-256

Edit: also the length should be at least 256 bits

It seems that meow and SHAMATA-256 are also this type of AES-NI hashing. They both got some Cryptanalysis, which seem not okay for our use case:

* [Cryptanalysis for MEOW](https://peter.website/meow-hash-cryptanalysis)

* [Cryptanalysis for SHAMATA-256](https://link.springer.com/content/pdf/10.1007/978-3-642-05445-7_1.pdf)

Edit: also the length should be at least 256 bits

But, i think falkhash is useful.

Btw, it is still recommended fletcher to use in rootfs, or, can we use blake3 for rootfs too?

It's not even in a stable release yet, so it's not really recommended for anyone to use for anything other than testing.

BLAKE3 should be tested currently. It has to settle down a bit and when it's considered stable - it will be put into some release also. Wether you use it for some rootfs is your decision.

@behlendorf - I think this issue can be closed, because falkhash should be defined first, and when this is done, it should have some cryptanalysis .... when this is done, it could become some new issue.