[ntp:questions] NTP architecture recommendation

cray74 dietmar.baur at gmail.com
Fri Aug 17 07:34:55 UTC 2007


On 17 Aug., 09:15, da... at ex.djwhome.demon.co.uk.invalid (David
Woolley) wrote:
> In article <1187269548.458817.315... at 22g2000hsm.googlegroups.com>,
>
> cray74 <dietmar.b... at gmail.com> wrote:
> > the offset between ALL servers needs to be < 3 ms; can't put GPS-
>
> What percentage of the time do you need to meet this requirement?
>
> > clocks in the datacenters (no GPS-signal); using free public NTP
>
> Use remote antennas for the GPS receivers; I don't think that you
> will achieve even 95 percentile 3ms spreads without it.
>
> Note that you will not be able to use Windows for any of your
> synchronised servers because time reading resolution for normal
> Windows applications is more like 16ms.  (ntpd on Windows may do
> better, but the applications will not get full benefit.)

Well, I need to clarify "datacenter": actually these are not OUR
datacenters, we use co-location at some provider, so we don't have a
lot of control there (some of them will not even allow us to feed
another cable into their server room). So even using a "remote
antenna" (on the roof) might not be possible in some areas. It really
looks like my approach will have to be: get a NTP-appliance with a
reference clock (like the Meinberg devices) as close to the servers in
the different sites as possible (ideally via a LAN).

Actually I will have to keep the servers in sync for 24 hours in the
near future (for now 12 hours is ok; we are in financial business and
will extend our operations to other markets soon, which basically will
extend our "business hours" to 23 hours a day).

And btw: we are not using any Windows servers :-).




More information about the questions mailing list