[ntp:questions] Re: An archetecture question....

John Pettitt jpp at cloudview.com
Thu Sep 29 17:01:51 UTC 2005


-----BEGIN PGP SIGNED MESSAGE-----
Hash: RIPEMD160

733t ASCII indeed :-)

You really need four servers to be reliable, they don't all need to be
accurate as ntp will figure out and ignore the inaccurate ones.  Four
will protect against one bad server.   If C is radically different
from A and B in your three server setup it will probably get dropped,
particularly if it's a lower stratum.

John
P.S. don't use prefer unless you really need to - ntp is pretty good
at working things out for itself.
P.P.S. for $80 you can add a Garmin GPS18-LVC and create your own
stratum one server (plus the cost of the host - figure $250 for a
soekris box) - if time is important to you this is the way to go.


Matt Kinard wrote:
> Okay, so here's the architecture we're looking at:
> 
> A redundant pair of stratum 2 servers (call them A and B) located in
> close phyiscal proximity to one another that will be serving time to a
> group of hosts across a high-speed WAN.  There is also an addional
> low-stratum server (call him C) located on another WAN that I could use
> as a back-up in the event of a communication loss to A and B (If I lose
> comms to one, I'll likely lose comms to both).  So what I've got is:
> 
> 
>                                              C------------------------
>                                                                                   
> |
>                                                                                   
> |
> A------\                                                                      
> |
> |            
> \                                                                     |
> |              | ------------------WAN-------------------Router/host
> |              /
> B-------/
> 
> Please admire my 733t ASCII skillz.
> 
> The configuration (on a Cisco Router) will look similar to:
> 
> /server A prefer
> server B prefer
> server C
> 
> /I know that this is a long way from an optimal configuration, but can
> you all see any glaring issues with the configuration?  Could you
> suggest a better alternative?  I've got a co-worker that is conviced
> that if C offers up a time that is radically different from A and B,
> we'll have problems.  Any thoughts?
> 
> 
> Thanks a million,
> 

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (MingW32)

iD8DBQFDPB3+aVyA7PElsKkRAyQHAKCPtpt2kgYALjnNfS+QEFdi/IW3rgCcC+TO
n3sEOKxOAHkfAk95YZ4cPhY=
=OIy5
-----END PGP SIGNATURE-----




More information about the questions mailing list