No CPU multithread ?

[ADMECA]

Hello, the application does not appear to be multithreaded.
When I encrypt large files, my CPU does not exceed 25% (Ryzen 7 5800H) is this planned for future versions ?
This would speed up the encryption a lot, right ?

ps: i use Windows version.

[HACKERALERT]

I'm not sure, I'm using standard go cryptography so depends on what they chose. There is hardware acceleration so I think it's fast enough already as long as you're not using RS.

[reimaj]

There is hardware acceleration so I think it's fast enough already

Fast enough for what purpose, though? I haven't looked into the implementation nor how the go libs do it, but I'd like to chime in with usage experience.

It's fast enough for smaller sizes(I'd consider <10GB small), but anything above that does take longer than any other tool available, which is why I sometimes resort to 7zip, but I'd really prefer to use Picocrypt for everything, with the more recent additions as well as the CLI it's the perfect portable tool for me.

I'd really appreciate a look into making the tool faster.

Maybe it's me using this tool for more than it was intended and I know encrypting big data was not the intention with this tool, but it basically already does that quite well. Besides, you've already added quite a lot of features, which you initially said you did not plan on doing and I think it has been a net positive for the tool. Besides, encrypting smaller sizes faster isn't exactly a net negative either.

[HACKERALERT]

ChaCha20 is not hardware accelerated like AES, so it's an inherent limitation and Picocrypt cannot go faster than 7-Zip. Does that extra bit a speed matter if the security is not as good though?

[hakavlad]

Does that extra bit a speed matter if the security is not as good though?

Multithreading will not reduce ChaCha's security.

The ChaCha20 algorithm can be accelerated through parallel processing of data chunks. ChaCha20 operates on fixed-size blocks of data (64 bytes), and its structure allows for the independent processing of multiple blocks.

Parallel processing can be implemented as follows:

Data Segmentation: The input data is divided into several chunks that can be processed independently.

Parallel Encryption: Each chunk is encrypted in a separate thread or process, allowing for multithreading to increase processing speed.

Result Assembly: After encrypting all chunks, the results are combined into a single output stream.

Sincerely, GPT4o-mini

[HACKERALERT]

Hacking together a bootleg golang.org/x/crypto wrapper to do that is probably not a good idea 😂

[hakavlad]

I assume that the BLAKE2 operation cannot be parallelized. Therefore, even if it is possible to parallelize ChaCha20, BLAKE2 will remain a bottleneck that does not allow increasing the overall speed by several times.

See also:

We specify 2 parallel hash functions (that is, with depth 2 and unlimited leaf length):
• BLAKE2bp runs 4 instances of BLAKE2b in parallel
• BLAKE2sp runs 8 instances of BLAKE2s in parallel

https://www.blake2.net/blake2.pdf