[time-nuts] New to Time Synching hardware - needing some advice
magnus at rubidium.dyndns.org
Sat Dec 29 00:07:47 UTC 2012
On 28/12/12 23:40, Ken Duffill wrote:
> On 27/12/12 23:22, Magnus Danielson wrote:
>> Sometimes when you have isolated machines, the absolute time may not
>> need to be very accurate, but the relative timing between them can be.
>> Adjusting from wrist-watch every once in a while may suffice for
>> absolute time.
> Yes, but what happens to the relative time when you adjust one of the
It takes time to propagate.
> The requirement seems to be to have a deviation of less than 10ms, but
> using a wristwatch is only going to give about a second resolution, so
> there could be a step change a couple of orders of magnitude bigger than
> that whilst the new time is propagated to subordinate machines.
> Surely, either don't bother (ever) to resynch to real time, or the
> problem remains. Wherever you get the master time from, you must be able
> to get it with a resolution of 10ms or better and you must measure how
> frequently you need to update the time so that the deviation is never
> greater than 10ms at resynch time.
You can have a real time requirement of the nodes in this isolated
system to be accurate to about a minute, but a relative timing within
the system to be within 10 ms. That is to say, it's OK that both node A
and B are 17 s off, if they agree with each other within 10 ms. Consider
TE_A = T(A) - T(UTC)
TE_B = T(B) - T(UTC)
TE_AB = T(A) - T(B)
If the isolated system only needs to agree within itself to be within
+/- 10 ms, then the difference of each of the nodes to proper time (UTC
in this case) can be way off mark. If it needs to be brought back to
reasonable errors (to compare logs to other logs), then either record
the time-difference or during maintenance window adjust the master of
those nodes, let the slave jump into place and then continue operations,
or at least log the adjustment event and expect the jump.
If you have a very sensitive and 24/7 operation, then just log the
difference with outside time and if needed adjust your log times with
the logged differences as you present the logs in offline processing.
There is nothing wrong with running an isolated system on it's own time,
it just creates the problem of correlating things in outside time, and
that creates the need for calibration logs and calibration measurements
(without adjusting). There is many ways to achieve the traceability
(legal or just layman).
More information about the time-nuts