Diving into the NTP pool (paper)

A while ago we (SIDN Labs) published a paper about the NTP pool.

It can be found here:

https://www.sidnlabs.nl/en/publications
(under ‘Diving into the NTP pool’).

Highlights:

(TL;DR)

  • Most clients are provided with fewer than 200 NTP servers, despite the fact that ~4.6k servers are listed on the NTP pool.
  • 10% of the clients see up to 12 IPv4 and 5 IPv6 NTP servers.
  • Most countries (60%) are served by 10 time providers only.
  • 13 countries are served by a single IPv4 time provider – which includes Israel, Pakistan, and Nigeria, while 42 countries are served by (the same) single IPv6 provider.
  • We found 9 IPv4 NTP servers on the NTP pool that deploy BGP anycast.

No shocking conclusions per se, but it might nevertheless be worth reading for those who are interested.

3 Likes

Interesting data.

I’m wondering if it would make sense to replace the system of country zones with something else. In large countries like the US clients get servers that are too far and in some smaller countries there are not enough servers, but there are close servers in other countries.

Ideally, the pool DNS server would be able to estimate the network distance between the client and each server in the pool and provide the closest servers, but that sounds difficult. As a much simpler model, maybe it could work with physical distance according to the location saved in the geo IP database?

I think admins could consider combining multiple low-coverage countries into a larger zone to increase the number of servers available.
One issue is that connection to a nearby country (especially in less-developed area) may actually be slower than connecting to a country farer away.
Maybe NTP Pool could incorporate server response time into the DNS system so servers are allocated not based on country zones but speed.

Who cares.

It is a rubbish protocol, it will never take over.

People do not get it, and it makes no sense at all.

In short, nobody cares about IPv6. :crazy_face:

Ipv6 is death unless they change it to something normal, at the moment it’s a stupid idea.

Don’t know how you come up with the conclusion that IPv6 is useless but this is not true lol

1 Like

Correct. I mentioned that in the very first sentence.

Not in this paper. We only made observations.

However, IPv6 is happening as we speak (can’t help it, sorry) and the NTP pool should indeed adapt to that development. Why it should, is extensively discussed here:

Only 30% of the websites can be reached over ipv6 only.
And the numbers of today do not show IPv6 will take over any time soon:

https://stats.labs.apnic.net/ipv6

Most people I know turn IPv6 off.

Capable doesn’t mean people use it.

It will never be dominant if they do not change the way it works.

IPv6 is growing. I guest you are not expecting an instant change like turning on a computer, right?

https://www.google.com/intl/en/ipv6/statistics.html

Btw, per the webpage which you shared, USA have nearly 50% and India have > 70% IPv6 adoption already :wink:

1 Like

Clients use it if it’s enabled.
That is not the point of IPv6.

The problem is at the server side.
Many systems do not use IPv6 because it’s soo damn complicated to setup.

For a client to do IPv6 or IPv4, most use dualstack, so the numbers of Google are not accurate except for Google themselves.

I really wanted to use IPv6 and tried for weeks, I got it working and then changed motherboard and I could change all DNS-entried.
Never got static working, it kept failing.

As so many others, I stopped trying as IPv4 works and is simple to setup.

That is my problem, why is it so stupidly complicated?

I’m not the only one, presumably more then half the world is too stupid to set it up, as very few websites enable IPv6.

More accurate numbers:

https://www.worldipv6launch.org/measurements/

70% can’t be reached over IPv6.

After more then 12 years, that isn’t a good sign. You would think they wonder why people don’t adopt it.

Yet they don’t. IPv6 will not take over, not now and not ever if they don’t make it less complicated.

Want to have more fun? The flow of traffic at AMX:

Total about 10TB!!: AMS-IX - Total Traffic

IPv6 traffic alone 500G!!: AMS-IX - sFlow Statistics

If you check the routers: AMS-IX - Routeserver Statistics

The numbers speak for themselves, after 12 years…

In short, the IPv6 traffic is near to none :rofl:

“IPv6 is complicated” > This sounds like an IT admin issue to me that they are reluctant to learn. However, for end users, many of them don’t even know that their devices are already using IPv6 as they just works.

“IPv6 adoption is slow” > As I have said, this will not happen overnight.

Some focus on the empty half of the glass while some focus on the gradually filling half.

1 Like

Sorry to hear that. And quite remarkable too, because it seems that all you had to do is disable SLAAC and/or RFC4941 privacy extensions.

Probably something like this (assuming Linux here):

sysctl -w net.ipv6.conf.eth0.accept_ra=0
sysctl -w net.ipv6.conf.eth0.use_tempaddr=0

(‘eth0’ is just an example, adjust at your own will and don’t forget to make it permanent, so it survives a reboot)

Then simply configure a static address just like you configured IPv4.

But what I really don’t understand:

After you failed; why won’t you accept that others are succeeding? Why do you want to prevent the pool from being reachable via IPv6 completely, so that all the many IPv6 clients can fully enjoy it?

How is it a problem to you if that happens? You don’t even have IPv6, so for you nothing will change.

That remains a mystery to me.

2 Likes

Some people are afraid to learn new things, it seems.
So in order to stay on top of the knowledge pyramid, the only option is to denigrate the new concepts unbeknown to them and to try and discourage the others from learning it, as they would soon be rendered useless otherwise.

A shame we’re still having this one-way “debate” year after year

2 Likes

Sorry? 70% refuses IPv6 today.

All you say is that others are succeeding? They are not, they face the same problems as I did.

THEY GAVE UP LIKE SO MANY OTHERS!

I tried to learn it, and all I learned is that nobody understands it, as it makes no sense.

IPv6 will never be the standard except for the lucky few that get it working.

Tell me this: If it was sooooooo simple, why didn’t it take over by the majority after 12 years? As it doesn’t.

Serverside IPv6 is crap, total crap. Why are you so dump to see that it’s not going anywhere, people like me just give up.

It’s a bad design, it’s a bad implementation, it lacks explenations on how to do it.

IPv6 will not take over this way, not now, not ever. The numbers prove it.

Even many IPv6 testing sites are outdated or simply do not work anymore.

It will not happen, simply it won’t.

12 years and counting…still not being adopted as usefull.

When will it take over, 50 years? 100 years?

If it was simple and working, it would have taken over years ago.

It didn’t. How hard is it to understand that it’s a bad design, bad implementation and badly executed.

In short, it sucks and therefor nobody uses it on purpose except a few idiots that where lucky it worked. :face_with_symbols_over_mouth:

Some 1500 brave volunteers spent time and effort to donate an IPv6 capable server to the pool (one wonders what message they are trying to convey to @ask with that :thinking:). That’s 48% of all NTP pool servers[1]. Fully functional and monitored, so in proven good working order. There are millions of clients out there ready to make use of them. And I don’t care if that’s 30% or 40% or 20% of the total population. It’s millions of clients.

Many of them already use 2.pool.ntp.org (the only one configured for IPv6 so far). But it could be a multitude more clients connecting via IPv6, if the other pool instances also would get an AAAA-record. I have already explained the benefits of that, so I won’t repeat myself here.

There’s a number reasons for that, sadly. But the alleged complexity of the protocol is not one of them.

[1] Based on the assumption that the vast majority of them are dual stack and also do IPv4.

This reminds me a joke to people who complain that Apple products are expensive: Apple products being expensive is not their, but YOUR shortcoming. :rofl:

Btw, you are contradicting yourself again : “nobody uses it” > This is wrong per the webpages which you shared.

If there’s anything I have learned over the years here, it’s that it’s totally pointless “debating” Bas on the subject of IPv6. He has a relentless need to post nonsense about it even in threads like this one which had nothing whatsoever to do with IPv6 before he brought it up. I think his father was killed by a 128-bit integer, and he has never gotten over it.

6 Likes

I think your conclusion is spot-on.

1 Like

It isn’t the adaption of the IPv6 being slow any more, we reached the top of the hill, the two protocols co-exist nicely.

From now on, it is the lack of phase-out of IPv4, which slows down the further progress. Weak up, please. If you agree, then you must pronounce this loudly on every platform.

We should put in place some plan for the phase-out of IPv4, at least in the context of NTPpool. (Think globally, act locally.) For example, switching the A and AAAA records in the geo-DNS code. Penalize the dinosaur IPv4 only users having 2.pool.ntp.org as the only available servers. (Sorry being so aggressive via exaggeration, I feel being obliged to shake the community.)

1 Like

The Pool is not powerful enough to push the change and this is not The Pool’s work.

It’s really up to tech giants e.g. Microsoft and Google to propel the change.
e.g. Microsoft Windows is still not accepting DNS assignment from IPv6 ND.

1 Like