[ntp:questions] ntpd, boot time, and hot plugging
per at hedeland.org
Wed Feb 2 23:55:41 UTC 2005
In article <mailman.12.1107379257.583.questions at lists.ntp.isc.org> Brad
Knowles <brad at stop.mail-abuse.org> writes:
>At 4:40 PM -0200 2005-02-02, Alain wrote:
>> 1) I am not saying of using ntpdate to *keep* a good sync but just
>> to guarantee a cold start, maybe just a little better then my whrist
>> watch, specially if I am not there.
> But that's not true. The problem is that ntpdate doesn't do any
>calculations to figure out if the server it's talking to is a
>truechimer or a falseticker -- it just does the query, takes the time
>it's given, and sets the clock. If the server you're connecting to
>has bogus time, then you have bogus time.
No, *that's* not true. Ntpdate sends four queries to each of the servers
given on the commandline, and applies "some" filtering to the results in
order to choose the "best" correction to the local clock. The algorithms
used are surely not as sophisticated as those used by (current) ntpd, but
then they don't need 15-30 seconds either (or 7-20, or whatever it is).
The decision to deprecate ntpdate and eventually drop it from the
reference implementation distribution has been discussed at length here
in the past, and whatever anyone may think about that decision, I'm
confident that plain ignorance about what ntpdate is and does wasn't a
factor when it was made.
>> 2) When you say "It does *not* give you good time sync" how bad is that?
>> does it test all the servers in the list using the NTP algorithm to
>> eliminate the bad ones? How much off can it get?
> See above. It can be off as much as they are off. Since UDP
>doesn't have any inherent security, it's trivially easy to spoof UDP
>packets, and trivially easy to do a man-in-the-middle attack. Heck,
>you can just continually spew garbage NTP packets with a bogus
>address, and if any clients come up in that timeframe, you've got a
>decent chance of getting them to believe you instead of the real
> With ntpd, you can at least require some sort of authentication
>key before you believe the server.
This diatribe is so far removed from reality that it doesn't really
deserve comment - let it be enough to note that ntpdate does support
"some sort of authentication", in case anyone wants to use it.
per at hedeland.org
More information about the questions