[time-nuts] Sun Blade 100 as GPS-controlled NTP server

[Chris Kuethe]

On Sat, Mar 15, 2008 at 9:26 PM, Matthew Smith <matt at smiffytech.com> wrote:
>  My first choice is my Sun Blade 100 as it is small, quiet, doesn't use...

last time I looked at a dmesg from one of those, it looked just like
my old hp laptop... but an ultrasparc cpu.

>  unless OpenBSD happens to have support for add-on USB PCI cards.

it does. i had an old machine with USB1.1 only. $10 at the computer
shop got me some super-cheap 2-port USB2 controller with an NEC chip
on it. worked like a charm. uou may have problems booting off an
add-on USB card... maybe the bios/firmware is smart enough to do that,
maybe not. once the kernel takes over you should be good to go.

>  The potential problem with the Blade is that it has absolutely lousy
>  timekeeping.  If run without NTPD, it will drift minutes out in a day.
>  I am advised that this is a hardware issue specific to that model.  I
>  believe that there is a patch to compensate for this in Solaris, but
>  it's seriously bad under Linux.

it's a pee-cee - of course its clock is lousy. solaris probably
compensates for this the same way openbsd would: calculate a frequency
correction over time, and adjust your clock frequency based on that.
let a userland program adjtime() fairly frequently to keep the clock
very close to right on.

if nothing else, can we get you to try it for a week and document just
how awful it is? (more than 5000ppm?)

>  The question is this - does it matter?  If I am running NTPD and feeding
>  in NMEA+PPS, is the native timekeeping (or lack of) an issue?

running on just the bog-standard crystal, my soekris seems to keep
time to within a few microseconds - i got similar results with 3
sunfire v120's. i think that's about as good as it gets without
expending serious effort to redo all the interrupt handling. adjtime()
gets called every 3 minutes or so to pull in those last few
microseconds, and a reasonable first guess at a frequency correction
will happen within 25 minutes. it'll take about 75 minutes for ntpd to
decide how to adjust those last few fractions of a ppm.

CK

-- 
GDB has a 'break' feature; why doesn't it have 'fix' too?