Kernel PLL, from IRC

[Gary E. Miller]

Yo Hal!

On Wed, 28 Sep 2016 11:14:49 -0700
Hal Murray <hmurray at megapathdsl.net> wrote:

> gemiller
>   I have a hard time believing a PLL
>   running at user timer interrupt time
>   can be anywhere as good as a kerenl
>   PLL running at TICK intervals.
> 
> There are two parts to this discussion.  There is the PPS time stamp
> and the PLL.

No.  PPS is unrelated to the kernel PPL.  It is a totally separate
kernel module.

> The PPS time stamp gives you the data to work with.

Yes.

> The work can be
> done in user space or in the kernel.

Yes, but not as well in user space.

>  New samples arrive every
> second.

No.  Twice a second.  But now you are confusing the PPS with the PLL.

>  Suppose I use the same algorithm.  Running the code in user
> space adds a few sub-ms kernel calls.

And a LOT more jitter.

>  (I'm assuming the scheduler
> cooperates.)

Bad assumption.

>  I don't think that minor delay will be significant.

Agreed, but the jitter is.  Also the frequency of updates is
different by orders of magnitude.

> The kernel PLL is probably using different parameters than ntpd.

As I have just seen.  Very dufferent results.

> gemiller
>   yes, best to measure
> 
> Check your clockstats on your minpoll-0 runs and see if it really was
> running at 0.

Yes, it was.  That is the recent bug fixes.

>  On my quick tries, it was actually running at 1 (2
> samples per batch) I didn't track it down.

Until Sep 26, then it got fixed.  But once again, unrelated to PPS or
PLL.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at rellim.com  Tel:+1 541 382 8588
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <http://lists.ntpsec.org/pipermail/devel/attachments/20160928/a0b77c6f/attachment.bin>