[ntp:hackers] NTP two-step on-wire protocol
David L. Mills
mills at udel.edu
Thu Jul 26 03:11:50 UTC 2007
Look carefully at PTP (IEEE1588). The on-wire computations are identical
to NTP. The leapsecond and TAI conventions are identical to NTP, except
that PTP runs on TAI, while NTP runs on UTC. The server selection
algoritm and transparent end-end and transparent peer-peer protocols are
not NTP, but the ideas could be ported over.
A PTP grandmaster is what NTP calls a primary server. A PTP boundary
clock is what NTP calls a secondary server. A PTP ordinary clock without
external source of synchronization is what NTP calls a client. Broadast
and client/server modes are in PTP, but not symmetric modes. The devil
is in the details.
I am not proposing NTP be ported to the Ethernet NIC, although at least
one 1588 Ethernet NIC has awesome horsepower with an onboard ARM prime
mover. I am promoting a scheme that allows PTP and NTP to interoperate
in the form of a modified NIC driver that can import and export
timestamps to and from the kernel system time variable. I am also
promoting output timestamps for NTP. Heck, let's just port FreeBSD to
the ARM and start a whole new adventure.
Tim Shoppa wrote:
> "todd glassey" <tglassey at earthlink.net> wrote:
>> Are we ass a group interested?
> Are you talking about embedding NTP into the stack which lives on
> the network card?
> Not me, I'm not interested. Why the heck would you want two different
> ntpd's running on a machine, one on the network card and one on the
> main CPU?
More information about the hackers