[time-nuts] GPS clock to RT-Linux for pulsar machine

Magnus Danielson cfmd at bredband.net
Sat Jul 2 06:17:21 EDT 2005


From: "Poul-Henning Kamp" <phk at phk.freebsd.dk>
Subject: Re: [time-nuts] GPS clock to RT-Linux for pulsar machine 
Date: Sat, 02 Jul 2005 08:47:48 +0200
Message-ID: <19366.1120286868 at phk.freebsd.dk>

> In message <000b01c57e96$7be31140$4023f304 at computer>, "Tom Van Baak" writes:
> 
> >PHK - do you have any anecdotes on how well NTP
> >works under primitive OS's like DOS?
> 
> "not at all"
> 
> DOS timekeeping is restricted to the 18.2Hz i8254 counter and that
> is far to crude for the NTP algorithms.
> 
> A few SNTP tools work and their precision is obviously 18.2Hz...

You can adjust the phase of the i8254 (or i8253 if you have a PC ;) just by
reseting it or stopping it. It's a bit crude, but on DOS you can own the CPU
completely during the process, so you should be able to phase adjust it with
some rather high accuracy. You can also read the i8254 counter. Since the i8254
is actually three independent counters where one is usually used for the
speaker, but you can disable that feature, then you could set up that counter
to run at a different frequency and then use their combined 32 bits as a phase
value (in need of some post-processing thought). So, if you really worked at it
you should be able to have the time adjusted or measured within a few us.

Only the adjustment period would require the CPUs full attention, but those
periods could be spread into micro-adjustments.

Besides, you can also change the 18,2 Hz into something more comfortable. :->

So, DOS do have its benefits in these cases, not alot of stuff between the
app and the hardware, so you can get quite alot out of it given some amount of
hacking. No wonder the Radio Astronomy guys still use it! The step from DOS to
RTLINUX isn't that strange after all.

Cheers,
Magnus




More information about the time-nuts mailing list