[ntp:questions] Syncing and peering for a multi-continent deployment
unruh at invalid.ca
Wed Apr 17 16:01:34 UTC 2013
On 2013-04-17, David Taylor <david-taylor at blueyonder.co.uk.invalid> wrote:
> On 10/04/2013 03:07, Blair Zajac wrote:
>> Thanks Charles,
>> OK. So unless I go for some GPS-based clocks in my servers, each data
>> center will use close stratum-2 servers and the stratum-3 clocks will
>> not peer with clocks in my other datacenters.
>> Regarding peering, then should each stratum-3 server in a single data
>> center use all 4-7 upstream clocks? Or should the 4-7 be split between
>> the three stratum-3 servers? The diagram at  suggests they should be
>> split? If that's the case, then how are falsetickers identified?
>>  http://www.ntp.org/ntpfaq/NTP-s-config-adv.htm#AEN3101
> That diagram (figure 5) shows each of the local prime servers being fed
> from just two external sources, but as you say, 4-7 would be more
> sensible. If it were me, I would go for one, two or three servers you
> know you can trust being in common across the stratum-3 servers, and
Also make sure that those link back to different stratum 1 or 0 sources.
It does little good to have 5 servers, all of whom derive their time
from the same stratum 1 source, when that common source goes down.
(ntpq -p tells you the source (refid) of the source you are using, and you can
follow that back).
Of course now adays, almost all refer back to a source that is GPS. So
even if they have different gps receivers they are all back to a common
source, the gps system. But perhaps if the whole gps goes down, or goes
bad, we as a society will have bigger problems than our computer clocks.
> fill in the rest with the "pool" directive. This will fill up the
> servers with up to 10 external sources, which are dropped and an
> alternative automatically selected if the source goes bad.
> I have information on my Web page about low-cost stratum-1 servers using
> Windows, FreeBSD and Raspberry Pi PCs.
More information about the questions