Client Polling Interval in NTP

[Harinath N]

Hi,

I want to make the client polling interval in NTP setup from 64 seconds to xxx seconds. I believe minpoll will do it, when addressed in /etc/ntp.conf. But even after adding the minpoll value in all our broadcast clients, still the ntpq -p in our client systems shows the poll interval as the same default 64 seconds.

Can anyone help me in this regard.
Also whether the NTP client polling will anyway load the network traffic?. Our site is with three subnets, comprising of 3 NTP servers along with 120 to 130 clients.

Thanks and Regards,
N.Harinath

[Sanjay_6]

Hi,

Did you restart the NTP daemon.

Thanks

[Sanjay_6]

Hi,

Here is the details for setting the poll time you are looking for.

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B2355-90147/B2355-90147_top.html&con=/hpux/onlinedocs/B2355-90147/00/00/63-con.html&toc=/hpux/onlinedocs/B2355-90147/00/00/63-toc.html&searchterms=Querying%20xntpd&queryid=20010930-220343

Hope this helps.

thanks

[Wim Rombauts]

Hello,

You are confucing me. How have you configured NTP ? With a broadcast server and broadcast clients ? Or with clienst that poll the server ?

In the first case, you should set the minpoll value in the "broadcast" statement of the NTP SERVER.
In the second cqse, you need to set the minpoll value in the "server" statement on th NTP CLIENT.

[Harinath N]

Hi Wim,

The NTP server is referring to its local system time with entry of "server 127.127.1.1" in its ntp.conf file and then broadcasting time to all other hosts in network using broadcast IP.

All the clients in the network work as broadcast client with an entry of "broadcastclient yes" in its ntp.conf file.

Now the requirement is, all the clients should poll the NTP server with a poll interval of XXX seconds. What should be done for this?

Thanks and Regards,
N.Harinath.

[Wim Rombauts]

About your global NTP configuration.

If you have no NTP server that synchronises to an external time server (on the internet) and if you have no clock to synchronise on, I would configure your 3 NTP servers to synchronise as peers with eachother and then broadcast the time on their subnet.
Then I would configure the NTP clients ad broadcast clients. This way, their will only be one broadcast signal every 64 to 1024 seconds instead of 30 to 40 polls requests + replies every 64 to 1024 seconds. Broadcasting reduces NTP traffic with a factor of 60 to 80, but can't cross subnets.

[Wim Rombauts · Accepted solution]

Normally, the clients don't need any further configuration. Just the line "broadcastclient yes" will make them listen for broadcasts, whenever there is one.

You just need to configure your server(s).
My guess for your ntp.conf file for your primary NTP server is :
server 127.127.1.1 prefer
broadcast minpoll 10

The NTP servers in your other subnet should have :
server prefer
server 127.127.1.1
broadcast minpoll 10

The second line in that file will cause the server to synchronise on itself in case he shuld loose contact with the primary NTP server. He will later resync with the primary NTP as soon as it is available again.

[John Payne_2]

ntp is a very low overhead protocal. You should not see much network load from ntp.

[Robert S. White]

Wim Rombauts is correct in his comments. The minpoll number of 10 will cause updates to occur every 1024 seconds (2 ** 10 = 1024).

The broadcast address should be the broadcast address of the subnet (example: 172.16.20.255 would broadcast the time to all hosts in the 172.16.20 subnet).

You must stop and start xntp. If /sbin/init.d/xntpd stop does not kill the daemon you will need to use the kill -9 command.

[Bill Hassell]

Why change poll rate? Are you concerned about the overhead? Typically, a single low end server (like an E35 or a 712) can handle more than 10,000 (that's 10k) NTP clients at the same time and the network traffic on a 10 Mbit LAN will not be noticeable. Here's the algorithm (not using boradcast, just simple polling):

64 bytes every 64 seconds per client until the client time is accurate to within 128 ms (that's 0.128 seconds) accuracy. Then the client doubles the polling time to once every 128 seconds, then 256, all the way to 1025 seconds (once every 17 minutes). On HP-UX boxes, when the /etc/rec.config.netdaemons file is set up with:

export NTPDATE_SERVER=12.34.56.78
export XNTPD=1

If you reboot, the time will be immediately changed before time-critical tasks like cron are started. From then on, the polling starts to slow down from 64 seconds.

[Sachin Patel]

Hi Hari,
Bill is right.

I have one NTP server 712/60HZ and it is serving 300 HP-UX systems. Now I am adding 100 Linux Cluster on it. I do not have problem from last 2 years.

Sachin