Re: Mac and NetBSD compatible usb to serial rs-232 (was: which serial-to-usb devices do you use?)

[Chuck Swiger <cswiger%mac.com@localhost>]

On Jan 15, 2011, at 10:42 AM, River Tarnell wrote:
> Chuck Swiger:
>> The FTDI USB-to-serial devices work well with both platforms (I was using 
>> them for a GPS receiver for NTP time-sync'ing purposes)
> 
> How well did this work for you?  I've always heard that for time sync use, 
> USB 
> serial adapters are to be avoided because the latency/jitter is much higher.  
> Unfortunately, many new systems no longer ship with real serial ports...

Excellent question.

The Mac version of a Garmin utility for the GPS receiver let me switch the unit 
to fixed location mode rather than mobile mode, which lets it average results 
over time for better positional accuracy and/or wanting fewer visible sats to 
keep a fix.  It seemed to figure out it's position and height above ground to 
the right building and floor level, both from work and at home.

I was able to get NMEA and PPS signals out of it via gpsd + SHM driver, so ntpd 
decided it was stratum-1.  The timing precision seemed to be comparable with 
the timestamps I was getting from nearby NTP timesources-- MIT and Columbia U. 
were both less than 20 ms network latency away from me, in NYC-- and all seemed 
to agree with offsets of less than 2-3 ms, and jitter generally under 1 ms.

I seem to recall trying to run at 9600 rather than 4800 baud, and playing 
around with different NMEA sentences, and while I could see reasonable results 
in tip or kermit (ie, # of visible sats, accuracy estimates), gpsd didn't seem 
to understand configs other than the default too well and would drop synch 
after a while.

I didn't have an oscilloscope and rubidium oscillator handy, which I'd need to 
try and judge timekeeping accuracy to better than millisecond level, but it 
seemed to be good enough to keep timestamps to the right scheduler tick at 
kern.hz=1000, maybe give-or-take 1.  On the other hand, running ntpd as a 
stratum-2 or -3, if you've chosen decent nearby timesources, seems to also 
provide that level of accuracy.  Folks who care about microsecond or nanosecond 
level timing would likely notice differences between network timesources, USB 
connectivity, and true serial or parallel connections for the PPS signal.

Regards,
-- 
-Chuck