Timeouts from San Jose

Server operators
1

Find attached (cant seem to attach them, please see link below) two log files both from the same server but one with IPv6 and the other with IPv4.

Can anyone explain what I should do to fix my IPv4 score?

You can see the summary of my whole system at http://www.ntppool.org/a/icolo/ you can see at that link that all the IPv6 is v stable but IPv4 is terrible. Many timeouts shown in the logs.

Looking at https://web.beta.grundclock.com/a/icolo I have much better scores, though those are from other monitoring servers.

2

As you described the scores depend on which monitor is used. The primary monitor in San Jose sees many timeouts for your IPv4 servers. (I’ve started some diagnostic traces on San Jose.) I checked one of my NTP clients located in London and saw near 100% response rate from all of your servers.

You’ll find many complaints about poor scores on community.ntppool.org , first concerning the Newark monitor (no longer active) and recently about the San Jose monitor. I wrote a note on the problem. Two Internet carriers, Zayo and Telia, are causing the most NTP loss. I have tried to get their attention for several years without success.

1 Like
3

I ran traceroute-like tests to your IPv4 servers for a day. It looks like the loss is in the Africa-> San Jose direction. We could go do additional monitoring, but it is unlikely to identify a work-around.

4

Hullo Steven,

Thanks so much for doing the tests. Its nice to know that my systems seem ok.

I guess this is part of the overall design problem of the NTP Pool with very few monitoring servers placed around the world.

I am in a position to offer free VMs and IP Transit out of my data centres here in East Africa for what its worth, but its up to the NTPPool team to take up that offer (I will make it again just now), and design a more distributed globals system.

My clients in the data centres clearly understand that they have to bring the content closer to the eyeballs, and so they provision Points of Presence around the world close to the consumer. This is established internet best practice which NTPPool needs to catch up on.

I am, in this case, about as far away as you can get in internet terms from San Diego, so no wonder the monitoring is poor.

Alex

1 Like
5

The .de capacity dropped by ~20% and my server was out of the pool for over a week.

See the flapping here: pool.ntp.org: NTP Servers in Germany, de.pool.ntp.org

de_zone

Personal statistics show a few things as well:

  • IPv6 is stable because no filters on the network route
  • Single San Jose monitor doesn’t work for a globally scaled service
  • San Jose monitor works even worse than Newark

newplot
newplot700×450 54.2 KB

1 Like
6

I reconfirmed my free offer to the pool leadership to provide infrastructure to additional monitors. Hopefully it works out for them.

My servers are basically never in the pool, even at 512 or 1meg bandwidth.

7

Alex, you can @Ask all you want, @Ask will never fix the pool or make better algorithms to fix the monitor.
We have been asking for years and years now. Nothing happens.
If you ask me; @Ask isn’t interrested one bit in his own project at all.

Good luck and pray a lot that something will be altered to make it better, so far it never happened.

8

My server has been dropped by San Jose for over a month…

Seems like some of the US telecom operators dropped all NTP packets from oversea on gateway