While the performance via IPv4 endpoints for the above mentioned pool members has been excellent, IPv6 performance has been spotty to say the least. Both of the servers are Digital Ocean droplets based in Bengaluru. Monitoring iptraf indicates that most IPv6 packets are coming through in large bursts in intervals of varying length.
What I found most interesting about this, is that the inblr2-1a6a7hp monitoring station does not seem have issues with the servers, and thus the problem might be more limited to international routes. Mostly for the sake of curiosity, and since I am new to serving in India, does anybody know what might cause such issues?
Those bursts probably correlate with one of the pool.ntp.org DNS servers returning one of your IPv6 addresses to clients. I’m guessing the vast majority of NTP traffic to the pool comes from poorly-written SNTP clients, as opposed to full NTP daemons like ntpd, Chrony, and openntpd, which at most will burst initially 6 or 8 times each 2s apart, or if misconfigured burst every poll interval 6 or 8 packets every 2s. On the other hand, simplistic clients have been known to do things like send continuously at short intervals until they get a response they can use.
It’s easy to believe there’s often congestion on relatively expensive international links from a country with a lot of poverty and very low-cost internet widely available.
That seems like a reasonable explanation, the reason it surprised me is that I have been running NTP pool servers in Europe (including IPv6) for years, and I have not seen this pattern before. Though to be fair, I have not had a reason to look for it either.
That is very interesting - which makes a lot of sense, especially since reducing the net speed setting seems to alleviate the issue.