Problem with GPSD refclock

[Rich Wales]

I reconfigured / rebuilt ntpsec with refclocks enabled, and everything
appears to be working fine now.

I'm using the "gpsd" refclock, and contrary to some other people's
experiences, it is running extremely stably for me, with jitter values
(from the peerstats file) consistently 10 microseconds or less, and with
an offset generally within +/- 2 milliseconds vs. several outside servers.

My Garmin GPS unit is connected via an old-fashioned RS232 serial port
powered by a 16550 UART built into my computer's motherboard.  When I
first wired up a GPS in 2009 for use as an NTP refclock, my
understanding was that USB-based devices (including "serial port" PCIe
cards which really used USB technology under the hood) were simply not
capable of the exact timing necessary to handle a PPS signal from a GPS
unit.  Even a serial port using a 16550 UART still required a special
low-latency driver / "line discipline" to handle the PPS signal promptly
and minimize jitter.  Perhaps things have changed in this respect since
2009 -- I really haven't investigated the situation in the past few
years, since my existing solution (even on newer mobos) continues to
work solidly.

Are the above concerns about USB-connected GPS/PPS hardware still
relevant today?  Many users today probably have no choice but to use a
USB-connected GPS -- and even I will, I assume, eventually need to do
this when my current GPS unit dies and I am no longer able to find a
motherboard with builtin 16550-based serial port capability.

Rich Wales
richw at richw.org