Performance

[Mikotochan]

https://twitter.com/hashbreaker/status/1023966188949463046

[jan-wassenberg]

Thanks for the link!

FYI our measurements are on Haswell (not Westmere), from real-world consumers of random bits (shuffle/sample). These have different performance characteristics than randombytes() into a large buffer.

Also, we have heard from library users that prefer smaller buffers (<= 256 bytes). There will always be a tradeoff between buffer size and cpb.

The paper discusses this in more depth. We expect to have it online in early-mid September.

[jan-wassenberg]

The paper is online: https://arxiv.org/abs/1810.02227

Questions/comments and discussion are very welcome.