Given that the probability of any leap events in the near future is currently rather low, and as it has been decided to generally do away with leap seconds anyhow, I donât see this as critical anymore nowadays. But if given a choice, other servers might anyhow be better suited, e.g., Appleâs. Or rather, diversity is what one should aim for, i.e., not too many eggs in/from the same basket.
Both Apple servers as well as Google servers are used as references for the pool monitors, by the way.
The pool indeed discourages having the pool itself as upstream, but not because âit gives you more cloudflareâ. Rather, due to the random nature of which servers one might end up upon each restart of the daemon, or when the daemon otherwise rotates servers learned via the pool (the NTP pool just has too many servers in at least some zones; other pools, such as Ubuntuâs, are less critical in that respect).
I.e., one approach could be to start using the pool initially, just to discover some servers that the daemon finds suitable. And then manually configure those servers specifically. (Or just do DNS queries for the pool names, and select some manually, and repeat if any of them does not perform well.)
Not more significantly than with other sources I would say.
Sure, diversity is usually better.
Yes. With emphasis on âdifferentâ rather than âmoreâ.
Not sure what you mean. Thereâs a maximum of four Cloudflare servers that a client could pick up. Not optimal, more diversity would be better obviously. But if the only âgoodâ servers you have access to in your zone are those from Cloudflare, what is one to do?
Sure, in some zones, I donât know why Cloudflare needs to be present. E.g., I had been thinking about raising a discussion about dropping Cloudflare from, e.g., the DE zone. But then, based on the numbers you say you donât understand, I realized their share in the DE zone is too low anyway to even warrant having a large discussion around it.
In other zones, they are the lifeline of the zone.
And then there are zones where Cloudflare servers are the only ones, so there, it does not make sense to have them at all from my point of view, as the implicit fall-back to the enclosing zone would/should take care of there being no servers in that zone (apart from the Cloudflare ones).
Though, maybe there used to be other servers in those zones that needed that âprotectionâ in the past (like your BE server is now operating under the cover of bigger servers recently added to the zone). Or it could help bootstrap an ecosystem of smaller servers which on their own would not be able to survive in a specific zone.
NoneâŚYou could run a server with only S6 servers as a source if you want (putting your server at S7)
Ideally you have a minimum of 3 servers so that at least 2 would agree closely and 1 could be detected as a âfalse-tickerâ. After that any odd number like 5 or 7 etc. servers would be preferred, just so that you always have a quorum of more than half of your configured servers agreeing on a common view of âcorrectâ time.
If you are able add as many low stratum servers as possible without resorting to higher stratum servers because a) the time precision doesnât get better at the end of a chain of âchinese whisperâ and b) many higher stratum servers may use the same reference clocks/S1 server anyway and you have no redundancy, just the same time from the same source diluted by errors through the chain of servers.
I would advise you to try and pick like 5 nearby low stratum servers from national laboratories or universities with low network latency. Just please check their websites if they are open access or make arrangements to be able to use them.
I know itâs just an example, but important to note that the pool will not accept server with stratum 7, and Iâm not sure off the top of my head as to whether 6 would still be accepted.
