[ntp:questions] Re: Mother Of All Clocks
hg at heiko-gerstung.de
Mon Mar 28 09:10:57 UTC 2005
Todd Knarr <tknarr at silverglass.org> wrote in message news:<z3D1e.19692$le4.1774 at fed1read04>...
> In comp.protocols.time.ntp <88e50c98.0503261447.2849adc7 at posting.google.com> Heiko Gerstung <hg at heiko-gerstung.de> wrote:
> > several millions of transactions per day. It is simply necessary for
> > them to ensure that all their networks (which are physically separated
> > due to security reasons) including all database servers, workstations
> > and routers are running with the same time coming from the same clock.
> That wouldn't completely explain the need for that many stratum-1
> servers, though. The only way I can see a need for 30+ stratum-1 sources
> is if the timestamp requirements were so stringent that you _had_ to
> have all clients sync'd directly to a stratum-1 to meet them and you had
> so many clients that you needed that many servers to avoid overload. I'd
> also want the stratum-1s physically seperated so loss of power or damage
> to a rack didn't take down all my time sources at once.
They have 16 physically separated networks meeting each other in this
one server room and yes, their time stamp requirements are _very_
stringent. Of course there is a protection against the loss of power
(UPS) and other problems.
> Normally I'd have expected a dozen or so stratum-1s scattered around the
> country syncing with each other (over their own dedicated network in
> this case), and a number (depending on the number of clients) of
> stratum-2 servers for each network that needs time service multi-homed
> between the stratum-1 network and the network they serve and syncing to
> non-overlapping subsets of the stratum-1s and sibling stratum-2s. I'd
> expect that to give you better resistance to failure (short of failure
> of the GPS system itself) and be cheaper as well. The only issue would
> be time stability, which I'd expect to be on the order of no more than a
> millisecond variation between the stratum-2s in this setup.
As I explained, their networks are located in one building complex, so
it would not make much sense to use 16 single stratum 1 servers (and
it would definetly lead to higher costs!). And, the policy of this
organization prohibits any physical network connections, they fear
that a bug in the OS or whatever could be used to enabled routing
where it should not happen, so connection the stratum 1's in a
dedicated time server network and then connect the several productive
networks to a node in this time server network would be against their
policy (note: it is useless to discuss whether this policy is actually
making sense or not, we are not in the position to change it).
The protection against a failure of the GPS comes from the hybrid
reception system, which includes a DCF77 correlation receiver (the
german radio time system based on an AM transmitter).
More information about the questions