[rescue] Time Servers

[Chris Kennedy]

Michael Dombrowski wrote:

> I've done some looking into this too. It looks like all you need is a 
> cheap GPS with NMEA serial output to get pretty darn accurate time, 
> down to about 10ms. This is fine for me, I don't know exactly what 
> you guys are trying to do and homework doesn't allow me to Google for 
> PPS right now.

I run two public access stratum-two servers (last check we had > 10K
machines from the UK to Sri Lanka slaving), which use among their
references a private stratum-one server based on a Trimble Accutime
2000.

Prior to picking up the Accutime I actually considered hacking together
a GPS reference, but gave up pretty quickly on using the NEMA output
of a generic GPS receiver.  While the lack of a PPS output can 
result in horrible jitter, the real problem is that the device
has (relatively speaking) lousy accuracy compared to a GPS time
source. 

The reason for this is that the generic GPS receiver's firmware is 
solving for position and does it's best to guestimate
clock inaccuracies introduced by SA and atmospheric conditions.
The Accutime 2000 (and most other GPS time sources as far as
I can tell) overdetermine their position during coldstart -- I
had mine take 150,000 position measurements to fix itself in
space.  Having fixed themselves in space, they can in essence
not solve for position and velocity, but for time.  

FWIW, I'm very happy with the Accutime; it cost about a kilobuck,
but I was in a geeky mood.  It annoys me that the nifty monitoring
and management software was Windoze specific, so I cobbled together
a Java app that does more or less the same thing and dumps
jitter stats as a side effect.  The only annoyance I had was the
fact that the Solaris driver for the thing tickled a bug in the
Solaris 2.8 serial I/O ioctls which required a minor hack to
make better.
--
Chris Kennedy
chris at mainecoon.com
http://www.mainecoon.com
PGP fingerprint: 4E99 10B6 7253 B048 6685  6CBC 55E1 20A3 108D AB97