This is a little rambly, but it’s also nothing urgent, so feel free to move on if you’re busy.
India is quite underserved
I have had a bare metal server at OVH in the Brazil pool for a couple years now, currently at the 1000 Mbps setting, and it’s getting a couple thousand queries per second.
I decided to put an old Amazon Lightsail instance in Mumbai in the pool, and once the score hit the cutoff to be in the pool, it was immediately seeing a flood of queries, at one point more than 10k a second. iostats shows an average of 1546/second.
This isn’t a big problem for me; I intentionally put it in to soak up some traffic. It’s just surprising that my busiest server isn’t the bare metal server in the Brazil pool at 1Gb; it’s a $5/month VM still at the 1Mbps pool setting.
If anyone else is weird enough to enjoy operating pool servers in underserved regions, India’s a great choice!
time.apple.com is surprisingly good
But it looks like time.apple.com resolves to 5 or 6 stratum 1 servers, and that it’s geo-based. Here in Boston, I get NYC, Ashburn, and Chicago servers of theirs. In India, it resolves to Hong Kong and Singapore stratum 1 clocks. For giggles I tried setting it up as
pool time.apple.com for a bit, and ntpd was very happy with them all.
Amazon Time Sync is good, but puts me at stratum 4
On my Lightsail instance, I configured 169.254.169.123 as a source. I have the “prefer” keyword configured for it, but I suspect ntpd may use it even if I didn’t. It is consistently the most stable clock, with sub-ms latency and even less jitter.
From the press release when they rolled this out a year ago or so, it sounded like they had some hardware stratum 1 servers in each data center, but they aren’t directly accessible, and they instead fan it out behind the scenes and make it accessible over a link-local address to VMs. 169.254.169.123 is present on AWS instances as an available NTP server, but it’s stratum 3.
The “problem” is that, although it’s the best source, using it puts my server at stratum 4. Worse, when I launched a Digital Ocean instance and got it to peer with the Lightsail instance, it picked Lightsail as the best clock, putting the DO box at stratum 5.
It’s not clear that this is a real problem, though. NTP is clearly smart enough to look at more than stratum. It’s just one of those silly things that irks me more than it should.