NTP support available?

[ldoolitt at recycle.lbl.gov]

Mihai -

On Wed, Sep 13, 2006 at 04:59:00AM -0400, Mihai Buha wrote:
> About the protocols:

Nice summary.

> * RFC 4330 defines the "Simple Network Time Protocol" (SNTP, a subset of
> NTP), on udp port 123.

I haven't read this one.  That's actually funny, since you (probably
accurately) describe my ntpclient program as implementing it.
I'll read it on the way to work today.

> * ntpclient implements a SNTP client. It just queries a NTP server, sets
> the rtc with the answer and exits. It is sort of a rdate using SNTP. No
> clock frequency adjustments, no detection of "outlier" servers. You must
> call it regularly to keep your clock in sync.

The first versions of ntpclient had no long-term lock capability.
It does now.  Just leave it running (with the right switches), and
it will phase lock, just like xntpd.  Still requires a single server,
but knows how to ignore outlier packets -- better than xntpd even.
Runs best if the machine's initial frequency offset is (approximately)
zeroed out, as described in its readme.

> My opinion is that in busybox, rdate should be replaced by ntpclient.

rdate should not be removed, it's different functionality.
If you show a compelling size advantage to ntpclient in busybox,
vs. ntpclient outside busybox, I would support the "port".
Note that we could end up with conflicts like we saw in
udhcpd, where the upstream author maintains a standalone
version, and there is a tension between maintaining a coherent
code base and conforming to the busybox standards.

   - Larry