Performance numbers

[Hal Murray]

These are the best numbers I can get - downhill with a tailwind.
The server is running with no-FUZZ and monitoring enabled.
I'd expect a few more cache faults if the client IP address is random.

Things are generally repeatable within a percent or so.  Sometimes, things are 
slower by 10% to 20%.  I don't know why.  I've seen at least one example where 
the slow-down was closer to a factor of two.

I'm preloading the pipeline with several packets (like ping -l) and adjusting 
the number of them for best results.  No packets are lost.

The first 2 systems on the list have Intel's AES-NI instructions.

(This might look better in a fixed pitch font.)

Microseconds per packet.
(or ns if you treat the dot as a comma.)
    NTP      AES      NTS
  4.614    5.713   22.302    Intel i7-4790 3.60GHz
  6.050    7.759   25.934    Intel i5-3570 3.40GHz
 27.888   42.556  212.308    Pi 3B
 25.358   35.938  169.748    Pi 3B, 64 bit
 15.249   24.856  105.910    Pi 4B
 37.938   62.707  334.422    Linux, 32 bit, Atom 230 1.60GHz
 36.507   59.759  259.442    NetBSD, 32 bit, Atom 230 1.60GHz
 40.900   65.727  355.343    FreeBSD, 32 bit, Atom 230 1.60GHz

Packets per second.
    NTP      AES      NTS
 216748   175047    45223    Intel i7-4790 3.60GHz
 165299   128886    38559    Intel i5-3570 3.40GHz
  35857    23498     4710    Pi 3B
  39435    27825     5891    Pi 3B, 64 bit
  65577    40247     9441    Pi 4B
  26359    15947     2990    Linux, 32 bit, Atom 230 1.60GHz
  27391    16737     3854    NetBSD, 32 bit, Atom 230 1.60GHz
  24530    15214     2814    FreeBSD, 32 bit, Atom 230 1.60GHz


-- 
These are my opinions.  I hate spam.