[ntp:questions] Re: IEEE 1588 support in NTP?
martin.burnicki at meinberg.de
Thu Nov 3 09:57:51 UTC 2005
Ulrich Windl wrote:
> martin.burnicki at meinberg.de (Martin Burnicki) writes:
>> You must also be aware that the timing accuracy is significantly degraded
>> if the PTP packets go through switches or routers which don't have a
>> special handling for such packets, so packets may be queued which
>> introduces again an
> It would be an interesting experiment to echo those PTP packets
> 1) directly via an X-cable
> 2) via some switch, hub, etc.
Of course the PTP folks have run such tests. Here are some values (from my
memory) using 100MBit lines:
Directly via X-cable you have only the cable delay, which is constant and
can be measured very good using harware TSUs. Accuracy can be a few 10s of
Hubs introduce a constant delay which is typically less than 1 microsecond.
Since the delay is constant, it can be compensated very good, and the
resulting accuracy can be in the same range as with a direct connection.
Switches introcuce a much higher delay, e.g. about 125 microseconds
(depending on the brand, etc.), as long as they don't have to queue the
packets. If they do have to queue the packets, the delay can be up to
several milliseconds, depending on the queue length.
There are special switches which account for PTP packets by implementing a
PTP boundary or transparent clock which cares about those delays.
More information about the questions