[ntp:questions] Setting up a NTP Time Server
unruh-spam at physics.ubc.ca
Tue Apr 8 18:57:06 UTC 2008
hal-usenet at ip-64-139-1-69.sjc.megapath.net (Hal Murray) writes:
>>Can someone point me to a good document that
>>shows how to setup a Time Server? I have an
>>isolated network that cannot get to the Internet
>>to sync time. I have Solaris 8,9,10 and Red Hat Linux
>>Advanced Server 4 servers as potential time server
>I don't know of any document that covers this case.
>It does come up occasionally. There might be some
>ideas in the newsgroup archives.
>I think your top level decision is do you want to setup
>a refclock or do you want to coast between manual corrections.
>For under $100 you can rig up a GPS clock. The Garmin
>GPS 18 LVC is usual suggestion. That will give you good
>time. I don't know anything about Solaris. Linux needs
>a serious patch to get PPS support. Without PPS support,
>your time won't be "good". It will be much better than
>coasting without any refclocks.
If you want I can send you the "Kludge" I use. I have the Garmin pps hooked
to the parallel port. I have an interrupt reading routine which puts the
timestamps of the pps input into /dev/gpsint. I altered the shm refclock
driver to read from /dev/gpsint. It works well.
>If you don't use a refclock, you can set things up
>to distribute the local system clock.
>It will drift between (manual) corrections. You can
>minimize that drift by calibrating your clock drift
>and putting the right value in the drift file.
chrony will also do that for you. Ie, you enter the manual times, and it
calibrates the cpu clock for you from those inputs.
>If your temperature and workload are reasonably stable,
>I'd expect you can get close to a second per week.
>A few odds and ends....
>The documentation for ntp is web pages, not man pages.
>There are generally 2 versions of ntp available: stable and dev.
>(There are also older version floating around too.)
>The dev version has new features, like the orphan mode. Maybe
>it has new bugs too. If you find web page via google, you
>don't know which version it is refering to. If you can, use
>the documentation that comes with the version of ntp you are running.
>I'd suggest a pass through all of the web pages. Skim the stuff
>that doesn't look interesting. Go back and read the important stuff
>Linux 2.6 kernels have a bug in the tsc calibration code. That will
>confuse things if you try to manually calibrate the drift. (I'll say
>more if you get that far.)
>These are my opinions, not necessarily my employer's. I hate spam.
More information about the questions