[ntp:questions] Second attempt at GPS-18 based NTP server

unruh unruh at invalid.ca
Thu Feb 2 21:22:25 UTC 2012


On 2012-02-02, David Lord <snews at lordynet.org> wrote:
> unruh wrote:
>> On 2012-02-02, David J Taylor <david-taylor at blueyonder.co.uk.invalid> wrote:
>>>> On NetBSD-5 with ntpd 4.2.6p5 the default is 64s.
>>> I wonder what expertise they have to select polling for reference clocks? 
>>> BTW: Paul has FreeBSD.
>> 
>> The reason for longer polling is a) network load, and b) rate
>> optimality.
>> If you are running ntp in order that your computer clock be ticking at
>> exactly 1s/s, then longer polls are better (longer lever arm) if your
>> computer clock does not change rate (eg due to internal temperature
>> changes due to use of the computer) If, you want least deviationo of the
>> clock from UTC shorter polls are better. Since the primary ntp noise
>> source, variable network delays, is not there, and since the primary
>> noise source is interrupt servicing delays, shorter is probably better.
>> 
>>>> I've just added the "minpoll 4 maxpoll 4" back to my ntp.conf
>>>> as it could reduce the 30 usec offset blips I get when logs are
>>>> rotated.
>> 
>> Not sure what would cause 30us blips. Are interrupts really being
>> switched off by something for those lengths of time?
>
> It's the load from log rotation and cron jobs running at
> that time of night.

The load should not matter. The interrupt sevice routines take
precidence. Now it might be that the interrupt routines require paging
in, but I thought that IRQ servicing was kept in memory as much as
possible. 
IRQs have to be fast. 

And 30us is a HUGE time  (over 10,000 cpu instructions)


>
>
> David Lord



More information about the questions mailing list