[LinuxPPS] LinuxPPS + ntpd coockbook

[Felix Joussein]

Hi Hal,

I am well aware, that most of this has already been discussed within
that list, but I think that it can get quiet annoying to browse dozens
of mails within a webUI, if you simply want to set up a nano second
enabled ntpd server which, again, doubtless is the main purpose for
LinuxPPS, isn't it?

So for the time, when patching the kernel will be obsolete, as it will -
hopefully - soon be integrated into the main kernel, this platform
should remain as a summary on how to use LinuxPPS for applications and
how to write code, setting up on PPS.

Until now it was focused on how to patch a mainline kernel so it can
read a pps signal from a CD line of a serial connector.
Once the mainline kernel will have LinuxPPS, depending on the default
.config setting, all this will be obsolete. Then, the wiki should
concentrate on how you can use this PPS signal.
Mostly this will be ntpd.
That for a manual should be written, which explains:

1.) best environment to place the new server
2.) hardware prerequisites
3.) the ldattach issue - which is now well documented
4.) the glibc / timex.h issue (as long as it will remain an issue = we
will have to test distributions which have that problem), and to be on
the safe side, place a timex.h to download and explain where to
integrate it in the ntpd source code.
For now, this could be documented here:
http://wiki.enneenne.com/index.php/LinuxPPS_NTPD_support#Atom_PPS_Clock_Discipline_.28refclock_atom_-_driver_22.29
Simultaneously ask the maintainers from ntpd to integrated the new
timex.h into their source code, as it belongs rather to ntpd then to
LinuxPPS.

Also nice:
coding examples for c/c++/(java?)

Regards,

Felix




> This information has all been talked about on this list and in most cases in 
> more detail.  Unfortunately not all of this info has made it into the LinuxPPS 
> wiki.  For example the wiki does not have any information on the glibc 
> nanosecond issues that started with kernel 2.6.26.  This should be part of the 
> LinuxPPS NTPD support page since we know it has a significant impact on how 
> well things work.  The fix recommended in the above web page it probably the 
> best one since it avoids patching glibc and it should be part of the NTPD page 
> on the wiki. 
>
> There has also been a lot of stuff here about different CPUs and 
> timers/counters that are also not in the wiki.  But I am not sure if any of 
> this information belongs on the wiki.  However it would be nice if there were 
> a single place on the web where this was documented (even the above link is 
> missing a lot of info about this subject).  
>
> One thing that is missing from both the wiki and the above web page is 
> anything about fixing the convergence problem in newer (2.6.19 and later) 
> kernels.  The fix is so simple it should only take one small paragraph to 
> explain it.  Why not include this in the wiki someplace?
>
> In addition, I sent a patch to the list for the Motorola Oncore section of the 
> wiki with detailed configuration instructions for that refclock.  This pulled 
> together information from various sources including information posted on 
> Remco's blog, information from Reg and stuff that I learned among others and 
> none of that information is posted yet perhaps because it was too detailed(?).  
>
> So there is lots of stuff that could/should be in the wiki that is missing.
>
> The flip side to all of this is that at some point the LinuxPPS patch set 
> should be part of the kernel main line and the need for a LinuxPPS wiki should 
> go away.  But when that happens I am concerned that a lot of valuable 
> information that is posted there will be lost.  But perhaps the focus of the 
> wiki will change to be more of a general howto for PPS and NTP on Linux which 
> will be clearly needed.
>
> Hal
>
> _______________________________________________
> LinuxPPS mailing list
> LinuxPPS at ml.enneenne.com
> http://ml.enneenne.com/cgi-bin/mailman/listinfo/linuxpps
> Wiki: http://wiki.enneenne.com/index.php/LinuxPPS_support
>
>
>