State of the microserver HOWTO

[Eric S. Raymond]

Gary E. Miller <gem at rellim.com>:
> Yo Eric!
> 
> On Mon, 6 Jun 2016 19:54:22 -0400
> "Eric S. Raymond" <esr at thyrsus.com> wrote:
> 
> > 1. How does ordering the refclocks in that way reduce startup
> > glitches? As written this is a maddening, uninformative tease
> > equivalent to mumbling "magic happens here".  Please reply via a
> > config with a better header comment.
> 
> As recently discussed on devel at ntpsec.org

Assume I've never read that list, or anything else about NTP other than the
HOWTO itself.  Remember who we're teaching!

> "On startup ntpd will take the first time it gets to set the system clock.
> If this first time is an imprecise clock, say derived from NMEA, then
> ntpd may takes days to restabilize.
> 
> The first time ntpd acquires will tend to be the ones higher up in the
> file with the lowest maxpoll.
> 
> So to work around this ntpd glitch put your best time sources high in
> the ntp.conf file, with your shortest maxpoll and your worst one at the
> bottom with higher maxpolls."
> 
> > 2. What has maxpoll got to do with ARP delays? See above...
> 
> As recently discussed on devel at ntpsec.org
> But let me take a stab at it.
> 
> "The default APR timeout on Cisco switch gear may be as long as
> 4 hours.  On windows and Linux it may be as short as 60 seconds.
> 
> If the polling interval for a chimer is greater than 60 seconds (maxpoll 6+)
> then when ntpd sends a time request to a remote ntpd daemon the OS may
> be adding an ARP roundtrip to the process, delaying the return
> by that much extra time.  This convinces ntpd that the remote ntpd
> is further away, and has more jitter, than it actually does.
> 
> To prevent this glitch in ntpd behavior, be sure to use 'maxpoll 4' or
> 'maxpoll 5' on local servers and peers."

Complete config with improved header comment, please.  Having me edit in
stuff every time someone needs to correct or amplify an explanation
will not scale and *will* drive me bugfuck crazy.

> You can include the logging statements with the comment:
> 
> # You want this logging, it will be usefull later.
> 
> If you add the logging early, then you have the data when you figure
> out you want it.  If you wait until you want it then it is too late.

Fair enough.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 811 bytes
Desc: Digital signature
URL: <http://lists.ntpsec.org/pipermail/devel/attachments/20160606/75cbbfab/attachment.bin>