I just ordered a digital clock that sits on the network and uses NTP to periodically refresh its accuracy, and looked at the manual. There’s a setting for how frequently it goes out to an NTP server for an update, and the manual showed it set at ten minutes. That got me wondering - clocks like this are mass produced, and maybe most of them are talking to NIST’s timeserver every ten minutes. And there must be something like a hundred million PCs in the USA, many of which check some default timeserver on a daily basis. How much traffic does a timeserver get? The messages being passed back and forth are tiny, and it’s not as though they’re downloading a web page with images, but still, isn’t the traffic enormous?
There have been several notable cases of NTP servers being hit by poorly written clients, especially with consumer grade routers. (Both Netgear and DLink have been guilty.)
Some data on loads is in the article and its links. Ten minute updates is not bad (but unnecessary for this situation.) It’s the once a second ones to one hard-coded server that cause problems. Especially when the server wasn’t intended to be a server to 10s of thousands of devices.
So every one of those mass-produced clocks is taking up an IP address? Just to update the time now and then?
That would make some sense if such clocks could join your local home network and get a 192.168.x.x address.
I’d expect the vast majority of them to be behind NAT-capable routers, so they probably are getting a non-routable address and get their traffic added to the single IP address the customer gets from their ISP.
Obviously, this leaves out sites where there is in fact a public-routable network. This is most common for businesses, but I have a /24 at home (and have 114 active devices on it, so it isn’t being hogged for no reason).
At work, I operate a group of 3 geographically diverse NTP servers which operate at Stratum 2. And yes, they all have permission to access the Stratum 1 servers they’re using. NTP traffic is so small that it doesn’t even get counted by our NetFlow analyzer - so it’s under .0001% of our network volume. And that’s all of our servers and all of our customers who use either those servers or some external-to-us servers.
Also keep in mind that a lot of businesses don’t rely on ntp servers. We set up our ntp about a year ago. We ran an antenna up to the roof which pings a satallite for the time for our ntp appliance. It was fairly cheap to put in, though the engineering guys had to poke some holes to get the antenna line to the roof.
Getting our core and distribution switches on ntp is a wonderful thing. Makes troubleshooting much easier.
Slee
I second that your clock is unlikely to have an ISP-assigned address. That would require a higher tier ($) of service from your ISP.
ISPs are running out of IPv4 addresses. Some have started rolling out a shared address scheme similar to NAT for the low end subscribers. You don’'t even get your own IP address. (Which raises questions about copyright infringement lawsuit provability.) IPv6 is long overdue in coming.
My clock has a configurable ip address for the time server.
The same company, Time Machines, also sells a Stratum 1 timeserver for $300. I think it sounds neat and fun to have but the only justification would be “because I thought it sounded fun”. It uses GPS and requires an outdoor antenna. Sleestak, do you perhaps have something generically similar (though perhaps not such cheap ones)? It doesn’t ping satellites and it is an NTP server so you would be relying on NTP servers, just not somebody else’s.
Italics added.
Could you explain that sentence? I don’t follow.
Ours in an internal NTP server, all our devices point to it. The server uses the GPS satellites to get its time. It is a Spectracom device. A blurb from their website:
Ours is set to GPS only.
We pushed out a time client that came with the package to our servers/workstations. I programmed in the time server address on the switches/routers manually.
Slee
Regarding lawsuits and IP addresses.
There are several companies out there that sue downloaders of copyrighted material. They monitor torrent downloads, Emule, etc. Collect IP addresses of the downloaders, contact the ISP to find out who has been assigned that IP address, and then send nasty letters threatening to sue for big bucks and ask for a settlement of “only” a few thousand dollars to avoid the suit.
If several people are sharing the same IP address, then things get complicated. You might get The Letter even though it was someone else who downloaded the movie. How do you prove a negative? Especially for less money than defending a lawsuit. Does the ISP tell the suers that it’s useless to try and ID the individual? Would that even slow down these trolls? Etc.
First, it’s probably using SNTP (Simple NTP) rather than NTP. When you’re only a client and don’t need to propagate time to downstream clients, SNTP is the correct choice.
SNTP is a lot easier to respond to than NTP: you just send back the current time and a few stats. Done.
10 minutes is pretty silly; I used to use SNTP and poll once a day, on my computers, before we started using ActiveDirectory at work which makes it unnecessary. For a niftly little SNTP client for Windows, google Automachron by OneGuyCoding. Not affiliated, just a satisfied user since mid-1990’s when I learned about NTP by working on it. (I’m even cited by Mills in SNTP V4 RFC, my biggest claim to fame. Woot!)
Without knowing what the hardware is, this isn’t answerable. But as folks mention above, it’s a trivial amount of traffic, normally.
With NTP (versus SNTP), the client backs off as the local clock gets regulated into phase with the server, up to 8 hours between queries IIRC. That helps keep the load on the server very low. The more significant limit (for full NTP) is the number of relationships the server can maintain.
With SNTP, no relationship is maintained, and the response (as I noted above) is easy to encode and tiny. If a server does nothing but SNTP, and if all clients polled at 10 minute intervals, and if the server can handle 1000 requests per second (which even a pretty wimpy server should), that’s 36 million clients (ignoring clumping).
Oops, math fail. 36 million is for 10 hours, not 10 minutes, poll time. (doh)
That would be 600,000 clients polling at 10 minutes, for an average rate of 1000 polls per second.
I hope 10 minutes isn’t the default, even if it happens to be the value shown in the documentation. 10 hours is way good enough.
Sorry, I don’t understand. Most inexpensive ‘atomic’ clocks and watches simply listen to a radio frequency to set the time. NIST, and foreign countries, broadcast the time, and some other information on certain radio frequencies. The little computer in the clock/radio simply listens and resets the time appropriately.
Desktop computers have a different mechanism and may indeed contact a time server.
GPS also transmits the time signal.
Bob
There are also IP clocks that synchronize over the network using NTP, not the RF time signal nor GPS.
Not available in AUS. Although in some parts of the country you can pick up Japan.