[nycbug-talk] dealing with drift
Mon Feb 28 13:38:32 EST 2005
* Isaac Levy <ike at lesmuug.org> [2005-02-28 13:25]:
> Thanks Dan,
> On Feb 28, 2005, at 1:16 PM, Dan Langille wrote:
> >If your ntp service can't sync with an outside server for a day or
> >so, you shouldn't get much drift. Your cluster will all be at the
> >same time, because the server is still up.
> In all practicality, very true.
> The situation where I'm trying to implement this is a bit strange
> though; where boxes that go down could be physically taken offline for
> perhaps weeks at a time- or never return to the cluster, so I'd like to
> try to make it as brainless as possible- (but in the end, brainless
> moves can require extreme brainwork/time when they fail :) - /me
> teetering on the fence all over...
You can run ntpd -q (or ntpdate) to perform a one-time clock jump
before your run ntpd. ntpd will not sync up if you are more than
1000 seconds off the reference server. ntpd -q (or ntpdate) gives
you a chance to jump close to the reference server in one jump.
ntpd will then sync up with a reference server.
In my understanding, ntpd does not set the BIOS time. It normally
sets the OS time.
Always use UTC as the time source and set your timezone accordingly.
More information about the talk