gpsd-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gpsd-dev] [PATCH] HOWTO: calculate the average GPS offset


From: Andy Walls
Subject: Re: [gpsd-dev] [PATCH] HOWTO: calculate the average GPS offset
Date: Mon, 21 Oct 2013 21:14:24 -0400

Hi Gary:

On Mon, 2013-10-21 at 16:52 -0700, Gary E. Miller wrote:
> Yo Andy!
> 
> On Mon, 21 Oct 2013 19:43:10 -0400
> Andy Walls <address@hidden> wrote:
> 
> > Days seems like too long.
> 
> I deploy a lot of servers.  Some have good clocks, some not.  Last year
> I saw two servers ntpd could not converge on at all, after weeks.   That
> is when I converted to chronyd.  YMMV.

Wow. :)

> And if you server goes though busy periods that will also delay getting 
> a god sync.
> 
> > The serial stream jitters *alot*.
> 
> hehehe, yeah.
>  
> > Looking at the peerstats for the GPSD/Shared Memory clock, and
> > throwing out the initial 6000 seconds, I observed the following about
> > the offset:
> > 
> > Mean: -0.455
> > Min:    -0.495
> > Max:   -0.413
> 
> Is that PPS over USB or PPS over RS-232?  That is within what I see
> for PPS over USB, but I have seen better.  If that is PPS over RS-232 then
> something is wrong.

That measurement set was:

Ntpd disciplining the system clock only using network connected servers.
The PPS (PPS Clock Discipline Driver) and GPSD (Shared Memory Driver)
clocks were include as peers which were marked with "noselect" in
ntpd.conf.  They never contributed to the disciplining, a.k.a. open-loop
for those peers.

PPS came in from a GPIO via the linux pps-gpio module and /dev/pps0
(IIRC).

Gpsd only had access to the UART with the GPS NMEA strings, and not the
PPS signal. (see why below.)

So that +/- 41 ms is the serial stream + kernel tty driver + gpsd
program jitter.  It does actually include about +/- 1 msec of PPS vs.
disciplined system clock jitter, but that is dwarfed by the +/- 41
msecs.



gpsd-3.7 ntpshm.c:init_kernel_pps()'s algorithm will never associate my
GPIO /dev/pps0 with the serial port of the GPS, AFAICT, so it will never
use it.



Regards,
Andy

> RGDS
> GARY
> ---------------------------------------------------------------------------
> Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97701
>       address@hidden  Tel:+1(541)382-8588




reply via email to

[Prev in Thread] Current Thread [Next in Thread]