On Thursday 12 June 2003 10:29 am, Paul de Vrieze wrote: > On Thursday 12 June 2003 16:04, [EMAIL PROTECTED] wrote: > > On Thu, Jun 12, 2003 at 03:54:40PM +0200, Paul de Vrieze wrote: > > Content-Description: signed data > > > > > > My understanding is that a combination of the two works well - and > > that's reflected in the gentoo (and other distros) ntpd startup script: > > > > - ntpdate is a one-time setter that uses the protocol to determine > > the current time. > > > > - ntpd is a daemon that uses the protocol to keep your clock honest. > > > > If your clock is way off, ntpdate changes it to the correct time without > > complaint. If you start up ntpd and your clock is currently incorrect, > > it'll correct a small error or give up on a large error. > > > > So the startup scripts do something that seems pretty reasonable: they > > run ntpdate as you're booting, then start up ntpd. Seems to work pretty > > well. > > That is correct. At bootup time no applications are running that would > suffer from a time lapse, so then ntpdate is safe. Running ntpdate from a > cronjob though, is not.
Actually, ntpdate is capable of slewing instead of stepping if the shift is small enough. You can experiment with cron frequency to run it often enough that it never steps. Some machines, that's once a day, some every ten minutes. Personally, I prefer ntpdate once and then ntpd, like Gentoo does it. I had settled on that scheme long before I started using Gentoo. Also, ntpd is better because even if your clock is being corrected by slewing with ntpdate, there is never any drift calculations like ntpd uses to make it so that your clock keeps correct time all the time, even when it's not able to poll the sync source. (how's that for a run-on sentence?) -- Democracy is two wolves and a lamb voting on what to have for lunch. Liberty is a well-armed lamb contesting the vote. -- [EMAIL PROTECTED] mailing list
