On Sat, Jan 5, 2019 at 9:14 PM Eduardo A. Bustamante López <dual...@gmail.com> wrote:> > On Fri, Dec 28, 2018 at 10:24:50AM +0100, Ole Tange wrote: > (...) > > Patch attached. : > I applied the Salsa20 RNG patch (slightly modified due to the recent changes > in > variables.c, attached [1]) to the tip of `devel` > (89b3a79dd4643f210f8443856214d558572733a5) and ran a couple of tests, to > answer > the following questions: > > - Does the new RNG generate uniformly distributed numbers? (Yes) > - What is the performance impact (roughly 2X slower) > - Does it break any existing tests? (Yes, easy to fix) > > > 1. RNG distribution : > 2. Performance impact > > The new RNG does more work, and thus, it is expected to have a performance > impact when generating lots of random numbers. I tested 3 systems (2 amd64 > and 1 > armhf) and include the results below. : > | BASH_VERSION: 4.4.23(1)-release > | time: 3.705 : > | BASH_VERSION: 5.0.0(1)-rc1 > | time: 8.983
That is an unfair comparison. You need to compare 5.0.0(1)-rc1+patch with 5.0.0(1)-rc1 to see if the delay is caused by Salsa20. My testing says the delay is _not_ cause by that. /Ole
