ntpd, chrony [was: Re: [plug] Backimg up entire drive]

[Cameron Patrick]

On Mon, Jan 05, 2004 at 07:04:31PM +0800, James Devenish wrote:

| In message <20040105110002.GB25582 at patrick.wattle.id.au>
| on Mon, Jan 05, 2004 at 07:00:02PM +0800, Cameron Patrick wrote:
| > chrony
| 
| What are the differences between chrony and ntpd clones? (I've always
| just used the thing that is called `ntpd` under whatever operating
| system I am using.)

Chrony works properly even when you don't have a permanent network
connection to the time server.  It keeps track of how fast the RTC seems
to drift, allows for manual overrides, can be told when you're connected
or disconnected from a network[1] (and even allows you to configure some
servers to be used even when you're offline), and is generally nifty.
It's a trifle harder to set up than ntp though.

It also lets you see how different your clock is to the NTP time without
forcing a discontinuous "jump" in the system clock like ntpdate does.
e.g.:

chronyc> tracking
Reference ID    : 130.95.128.58 (elysium.uwa.edu.au)
Stratum         : 3
Ref time (UTC)  : Mon Jan  5 11:11:37 2004
System time     : 0.000000 seconds fast of NTP time
[...]

On Mon, Jan 05, 2004 at 07:11:48PM +0800, James Devenish wrote:

| I assume that's the OpenServer machine (no pgrep + ntpd that crashes)?

Ooh, cheers, I discovered a new command!

$ pgrep ntp
666

Heh.

In my experience ntp tends to die randomly if the system clock is really
really dodgy compared to the time server.  (c.f. the thread on this list
from a while ago, with a subject line something like "ntp exceeds
sanity limit")

Cameron.

[1] In the particular case of Debian and a PPP internet connection, this
is done automatically.  Otherwise some (trivial) shell scripting is
called for.