GETTING STARTED WITH NTP

by Tom Limoncelli
tal+hp@limoncelli.org

Step 0: Is it ntp or xntp?

Some operating systems supply binaries with the "x" and sometimes without the "x". This document uses options that should work for either. I'll always specify the command without the "x". If you receive a "command not found", try it with an "x".

Step 1: Pick your servers

Find some public servers that you can sync with. Two is good enough. There is a complete list here: http://www.ntp.org (it's a link to http://www.eecis.udel.edu/~mills/ntp/servers.html )

These sites are kind enough to perform this service for you, you should follow any guidelines they request (asking permission first, donations, thank-you notes, whatever.)

Let's suppose you've picked clock.example.org and timekeeper.sample.com. Those aren't real sites, but I'll use them as examples in the rest of this document.

Step 2: "Manually" set the clock correctly just once

Let's "manually" set the clock to the correct time just to start things off. When ntpd is running and finds your clock is wrong, it makes tiny little changes until the clock is right. This way applications don't get confused. However, on reboot, the clock may have lost a lot of time and since no applications are running yet, we can make a big timeleap to correct the clock. To do this, we use a different utility called "ntpdate". It can't run at the same time as ntpd. We even give it the "-b" option so that it knows to leap forward or backwards in time to set the clock correctly.

Here's the command that does this:

	ntpdate -b clock.example.org timekeeper.sample.com

(NOTE: Substitute the 2 NTP servers that you'd like to sync with.)

Try this once from the command line (as "root") to make sure you're got it right.

	# ntpdate -b clock.example.org timekeeper.sample.com
	 5 Jun 08:50:18 ntpdate[23777]: step time server 10.10.10.10 offset -111.106949 sec

This means that your clock was off by -111.106949 seconds, but now it's been brought up to date. The good news is that for an instant you were in sync. The bad news is that by the time you read this, your clock has drifted. Darn computers!

Step 2: Set the clock correctly on every boot up

On reboot, you want to tell the system to sync up quickly. While your machine was down the CPU gets bored and plays with the clock. Haven't you ever been in a clock store waiting for your mother to make her purchase and started setting the clocks to funny times? It's like that, only your mother has nothing to do with this (I'm sure she's a very nice person, you just don't appreciate her enough).

Let's configure your system to run this command on boot up. Here's how to do this on various operating systems:

FreeBSD:

Simply add this line to the end of /etc/rc.conf:

	ntpdate_enable="YES"
	ntpdate_flags="-b clock.example.org timekeeper.sample.com"

Solaris 9:

This is done automatically if a properly configured "/etc/inet/ntp.conf" exists.

(people can volunteer the easiest way to do this on their favorite OS)

Step 3: Keep the clock in sync permanently and continuously.

Having your clock set properly every time you boot is nice, if you reboot constantly. Of course, you could put that command in "cron" but we have a solution that is so much better you'll thank me in the morning. We'll run "ntpd" which will constantly keep your clock correct by making micro-adjustments all day long. Heck, ntpd is so smart that if you lose contact with all your NTP servers it will remember how bad your clock was and keep making adjustments based on past bad performance. Pretty cool, eh?

The NTP documentation has volumes about the various features and entire books can be written about the theory of operation of NTP. It's really quite amazing how the system works so precisely, so accurately, does the right thing through outages, detects and avoids misconfigured servers, conserves bandwidth, and fixes problems that you don't yet realize you have. However, you don't care about that, you just want a simple configuration that works. Here's one:

Create a configuration file. Different systems call it different things.

FreeBSD 4.x:: /etc/ntp.conf
Solaris 9:: /etc/inet/ntp.conf

Put these 3 lines in your configuration file:

	driftfile /etc/ntp.drift
	server clock.example.org
	server timekeeper.sample.com

Now start ntpd.

FreeBSD 4.x:

	/usr/sbin/ntpd -p /var/run/ntpd.pid

Solaris 9:

	/etc/rc2.d/S74xntpd start

It will now sync with the 'servers' in /etc/ntp.conf. Once an hour it will record a little information in /etc/ntp.drift. (/etc/ntp.drift contains what it needs to know if it loses contact with all your servers and needs to "go it alone").

Of course, you'll want to make sure that this command is run on bootup, after ntpdate is run (they can't run at the same time). Here's how to do that on various operating systems:

FreeBSD 4.x:

Simply add this line to the end of /etc/rc.conf:

	xntpd_enable="YES"

Then use "touch" to create some initial files:

	touch /etc/ntp.drift /var/log/ntpd.log

Solaris 9:

This is done automatically if a properly configured "/etc/inet/ntp.conf" exists.

Step 4: What about stratum 1, 2, 3 and all that?

You don't need to specify that in the configuration file. If you list a mixture of stratum 1s, 2s, and 3s ntpd will do the right thing. Your stratum is determined dynamically based on what you've synced to. You don't need to worry about this. However, if you want to know more, there is info in the documentation.

Step 5: Check your work.

"ntpdc" is a command that lets you talk to your ntpd (just like "lp" has "lpc"). It has a zillion features, but the one you really need to know is:

	ntpdc -c peers

Here's a sample run:

	> ntpdc -c peers
	     remote           local      st poll reach  delay   offset    disp
	=======================================================================
	=clock.example.o 5.0.0.0         16   64    0 0.00000  0.000000 0.00000
	=timekeeper.samp 5.0.0.0         16   64    0 0.00000  0.000000 0.00000

Output like that (all zeros) means that we haven't synced with those two servers yet.

When things get synced, the numbers will fill out like this:

	> ntpdc -c peers
	     remote           local     st poll reach  delay   offset    disp
	=======================================================================
	*clock.example.o 64.46.24.66    3   64  377 0.00018  0.000329 0.00255
	=timekeeper.samp 61.26.45.88    3   64  377 0.00017  0.002122 0.00342

st -- the stratum of that server (calculated automatically)
poll -- how often you poll that site (backs off dynamically)

ntpdc can talk to remote machines if you have the proper access set up. Just list the hostname on the command line:

	% ntpdc -c peers ntp-2.cso.uiuc.edu
	     remote           local      st poll reach  delay   offset    disp
	=======================================================================
	+ntp-1.gw.uiuc.e 0.0.0.0          2 1024  172 0.00627 -0.000044 0.00761
	+tick.usno.navy. 0.0.0.0          1 1024  373 0.09023 -0.023207 0.15714
	*truechimer.cso. 0.0.0.0          1 1024  377 0.00401 -0.000054 0.00006
	+ntp-0.gw.uiuc.e 0.0.0.0          2  512  332 0.01059  0.001216 0.00497

Step 6: Running NTP on many machines in a network

If you have a lot of machines that need to be kept in sync, it is rude to point them all at public NTP servers. It's also a waste of your internet bandwidth. A much better solution is to set up 1-2 NTP servers on your network that point to public NTP servers, then point all your other machines at these 1-2 NTP servers. This greatly reduces the amount of load you put on the public NTP servers.

At my company, we have 2 NTP servers: ntp1.mycompany.com and ntp2.mycompany.com. Those are actually aliases ("CNAME" in DNS) that point to the 2 machine that I want everyone to use for NTP service. We do another trick that makes the aliases point to different hosts outside our firewall. That way the same configuration file works if a machine is connected inside or outside our firewall.

These 2 NTP servers need to have a custom configuration, that includes each other:

	driftfile /etc/ntp.drift
	# public NTP servers we sync to:  (use these 2 lines ONLY on ntp1 and ntp2!)
	server clock.example.org
	server timekeeper.sample.com
	# Use these lines on all machines in mycompany.com:
	server ntp1.mycompany.com
	server ntp2.mycompany.com

I find that it is easier to have the exact same file on both of these servers. That means that ntp1 will try to sync with itself, and ntp2 will try to sync with itself. Luckily, this is ok. NTP will realize that its talking to itself and do the right thing (ignore itself).

All the other hosts have this ntp.conf:

	driftfile /etc/ntp.drift
	# Use these lines on all machines in mycompany.com:
	server ntp1.mycompany.com
	server ntp2.mycompany.com

Step 7: Other resources

http://www.ntp.org
Read the ntpd man page

-=end=-

Note: I make no endorsement of this product, but they asked for a link exchange so I obliged:

Galleon Systems - Time Synchronisation Specialist. Galleon provide NTP Servers for time synchronisation of computer networks.

I do, however, completely endorse this product for system administrators that want to be able to better manage their time:

O'Reilly's Time Management for System Administrators - NTP manages time for your computer. If YOU need help with HUMAN time management, this is the book for you!