[time-nuts] RS 232

[John Miles]

> john at miles.io said:
> > Agreed, nobody should be using RS232 for anything nowadays.
> 
> RS232 works much better for capturing PPS timing.

Unless you are watching it with a ring-0 (kernel) driver, and/or using a
hard realtime OS to run the client software, it really won't matter that
much.  Anyone running Windows or most flavors of Linux has more to worry
about than the distinction between USB and RS-232, when it comes to latency.

For truly critical applications it's best if the counter itself does the
timestamping.   For ordinary NTP use on Linux or Windows the distinction
between RS232 and USB is pretty questionable.  Submillisecond jitter has
been documented in USB PPS applications (e.g.,
https://lists.bufferbloat.net/pipermail/thumbgps-devel/2012-March/000109.htm
l ), albeit with unspecified latency.  If that's not good enough, you need
to tackle the issue somewhere besides the physical layer.

> Another advantage of RS232 over USB is that the configuration is stable
when
> things get unplugged and replugged, or powered off, or ...  Of course,
that's
> a disadvantage if your program wants to know when the gizmo got unplugged.

USB devices have gotten a bad reputation in this regard because of
developers' failure to understand the idea behind serial numbers.  As with
noise immunity, it's possible to do it right, it's just that too many people
don't bother.

-- john, KE5FX
Miles Design LLC