Doubt about server rotation in Pool DNS

Recently I’ve been checking the DNS records of the us.pool.ntp.org address and realized that one of my servers (with netspeed at full speed) constantly appears as record A on various DNS servers in North America.

However, I have a secondary server also located in the USA, but with netspeed at 25Mb/s. This server receives about 300 requests per second, and even after performing multiple checks on the us.pool.ntp.org and noth-america.pool.ntp.org domain, DNS servers have never returned the IP from that particular server.

This leads me to believe, that depending on the netspeed set, the server IP is not included in the regional pool rotation list. If NTP traffic is not coming from the us and north america pool, in which pool domain is this server being distributed?

Just out of curiosity; how did you check? Hopefully not against your caching local resolver, but instead directly at (one of) the authoritatives of pool.ntp.org?

The US zone has a large number of servers and many of them are configured at high speed. At the maximum 1Gb/s setting you would get only about 1% of the zone’s traffic. With the 25Mb/s speed you might need to make thousands of DNS requests to actually see your address.

1 Like

I checked using the https://dnschecker.org/ service which queries several recursive DNS servers around the world.

1 Like

I ran IPv4 DNS tests against one of the authoritative servers for [0123].us.pool.ntp.org
522 unique server addresses were returned, the number of servers in the zone is currently about 550. The distribution was highly skewed toward the high-rate servers, as it should be

2.67% 162.159.200.1 time.cloudflare.com
2.65% 162.159.200.123 time.cloudflare.com
0.008% My US-based pool server. Configured rate = 10Mbps.
It is over 600 times more likely for a Cloudflare server to be selected than my server.

2 Likes

The frequency of distribution of servers with low netspeed seems to be actually very small compared to servers with maximum netspeed. But judging by the amount of requests, it seems correct. We should also consider that many clients cache DNS results, ignoring new query to the pool zone.