I love the idea of a more friendly client - but perhaps the patch could be as
simple as doing a new name lookup if the current server is not responding?
So for example a client has configured:
server 0.pool.ntp.org
server 1.pool.ntp.org
server 2.pool.ntp.org
The machine he or she resolved from the name "1.pool.ntp.org" disappears for
more than {arbitrary long time period - eg 24h}. Now resolve 1.pool.ntp.org to
a new IP address and sync to it instead.
I can imagine that there are huge benefits in terms of time stability to
keeping the same servers active if they are responding with accurate time.
Dave.
--
David Rawling
Email: [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
________________________________
From: [EMAIL PROTECTED]
Sent: Fri 26/01/2007 7:50 AM
To: Guillaume Filion
Cc: [email protected]
Subject: Re: [time] NTP client 'fall-out'
On Thu, 25 Jan 2007, Guillaume Filion wrote:
> There doesn't seem to be a lot of activity from the people at ISC to
> make changes to ntpd to make it more pool friendly. I've been thinking
> for a while about making a patch for ntpd to do this. Linux distros
> could include our patch in their packages so a good proportion of the
> user base would be running this version in a couple months.
>
> I'll post to timekeepers-dev with more details.
>
Excellente suggestion !
Would you just patch ntp so it quits trying to connect to an ntp server
after so many attempts or time period ?
If so, what would be the numbers for that parameter ?
Then again, a down server might come back up at anytime, so obviously
ntpd as a client could stop trying to connect to the servers too soon.
What would ntp do when one server goes down? In the long run, it will need
to reconfigure a new one to replace it.
So here is a suggestion for the patch :
1) ntpd stops trying to connect to a server after so many attempts or a
given time period ( a time period could be better )
2) change ntpd configuration so we can add 'spare' servers that would not
be used when ntpd first starts up. ntpd would only start connecting to the
spare servers once one of its 'main' server would stop responding.
3) ntpd would use spare servers to replace main servers.
4) ntpd could also verify if one of its main server as come back on
line say, once every week and failback to it.
My main concern is that knowing that a server can be taken off from the
configuration, users of ntpd will be tempted to put many servers in their
ntp.conf file and that this will generate more useless ntp traffic. This
is why I introduce the concept of 'spare servers' to solve this.
Of course, you could implement a more pool specific solution ( just doing
a new DNS lookup at the pool when one server goes down ) but I would be in
favor of a more generic solution ( active and spare ntpd server in the
config )
Maybe there is a simpler solution so let us know what you plan to do.
actives and spares is a well known concept although.
-Louis
> Cheers,
> GFK's
> --
> Guillaume Filion, ing. jr
> PGP Key and more: http://guillaume.filion.org/
>
>
Louis
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
