[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [hobbit] ntpdate deprecated
- To: hobbit (at) hswn.dk
- Subject: Re: [hobbit] ntpdate deprecated
- From: Henrik "Størner" <henrik (at) hswn.dk>
- Date: Fri, 13 Nov 2009 10:51:26 +0000 (UTC)
- Newsgroups: lists.hobbit
- Organization: Linux Users Inc.
- References: <af5861d60911110300l2a45d732od87076f4be76ee7b (at) mail.gmail.com> <af5861d60911110300l2a45d732od87076f4be76ee7b (at) mail.gmail.com> <200911111352.18786.bgmilne (at) staff.telkomsa.net>
- User-agent: nn/6.7.3
In <200911111352.18786.bgmilne (at) staff.telkomsa.net> Buchan Milne <bgmilne (at) staff.telkomsa.net> writes:
>NTP is not that useful for monitoring the time on monitored hosts, (the
>client's built-in check is better, see the CLOCK option for hobbit-clients.cfg
>thresholds), unless you are running all your hosts as ntp servers.
Sorry, but I don't agree.
The CLOCK option works by comparing the timestamp a client puts into
the client-message with the timestamp on the Xymon server by the time
that client-message is received. This can easily vary several seconds,
depending on
- the load of the Xymon server
- network latencies between the client and the Xymon server
- the load on the client
- how long it takes the client to generate the client data
- how well time is synchronized on the Xymon server
- and probably something else
That's why the default threshold for the CLOCK test is 30 seconds;
it really is not a very accurate way of determining the actual
time difference between a client and the "real" time.
And for some types of servers (notably Windows domain controllers
and other Kerberos-style servers) it is completely inadequate.
You're right that it requires some code-changes to use the "sntp"
utility instead of "ntpdate". I'll merge some code for this into
4.3.0, that will also allow us to drop the old "ntpstat" extension
since we already have that data from either ntpdate or sntp.
Regards,
Henrik