Hi, On Fri, May 29, 2020 at 03:33:28PM -0700, Kevin J. McCarthy wrote: > On Fri, May 29, 2020 at 03:58:35PM -0400, Remco Rijnders wrote:
> I think your comments in <XsvMOtX+QDTV/kA0@settler> make a good case for not > putting crypto-level randomization into a PRNG function, and I would like to > think about it before moving in that direction (for message-id's too). Regarding crypto libraries an important reason to avoid it (if included in build anyway) is if performance is not competitive. I'd expect that the difference (impact of side-channel free coding) is not that we worry about it. As random numbers are required only for specific, not permanently occuring events, slight disadvantages probably are not noticeable. What are performance expectations and methods for measurement? Generate random numbers alternating with context switch (to not run purely CPU top-level caches) such as a kernel rwad/write calls? Gero
