On Thu, Nov 06, 2014 at 06:08:32AM -0500, Jared Mauch wrote:
> One thing I will highlight here is we have ips in both v4 and v6 pools that 
> are anycast and there may not be a good way for something to realize 4 ips 
> may just be 2 hosts with the same reference. 

So the problem is that a client could be unknowingly using the same
server twice, which would effectively give it two votes in the source
selection? Can this happen with addresses from one DNS reply or only
over time when some sources are replaced?

I think the client could theoretically detect that by comparing the
reference timestamp, reference ID, stratum and poll of the servers. In
order to compare reference timestamps from the same clock update, the
polling of the servers matching in the other fields would need to be
grouped first. This could be tricky to implement.

In the pool mode when there are many sources available, maybe it would
be acceptable to ignore the reference timestamp and replace sources
even if they are not actually duplicates of others.

-- 
Miroslav Lichvar
_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool

Reply via email to