[ntp:questions] Re: IEEE 1588 support in NTP?
David L. Mills
mills at udel.edu
Mon Oct 31 02:44:37 UTC 2005
I have seen the 1588 document. I hear newsflash from your competitor
Syummetricom on the same issues. However, affordable PTP interface
devices will need to be available for various networks (10/100/1000
Ethernets, FDDI, etc.) for the technology to propser.
While 1588 provides a precision timestamp, it doesn't say what to do
with them to discipline the system clock, if that is required. The NTP
clock discipline algorithm might be a good candidate for that. This is
not a trivial algorithm, along with kernel support, but would be
necessary if the overhead of frequent PTP exchanges becomes a problem.
An interesting project would be to write a driver for the PTP device
that passes the timestamps and data on to NTP and modifiy the protocol
module to use them. This needs to be done carefully, as the I/O cacheing
on Solaris, for example, seems to be ever increasing.
I suspect your product, like the Symetricom product, includes both 1588
and NTP. I would assume also that 1588 could serve as a reference clock
for NTP and possibly the other way around, like for instance to
synchronize time among 1588 islands.
Martin Burnicki wrote:
> Our company Meinberg is offering a NTP time server which also works as a PTP
> grandmaster clock with a built-in hardware time stamp unit. We've recently
> participated in the IEEE-1588 meeting in Winterthur near Zurich/Switzerland.
> Part of that meeting was a plug-fest where PTP capable devices from different
> manufacturers were interconnected in order to see that they could work
> together. For more information please see our news article:
> David D. Hagood wrote:
>>Has any consideration been given to adding support for IEEE-1588
>>Precision Time Protocol (PTP) into NTP?
>>PTP (http://ieee1588.nist.gov/) is a standard for time synchronization
>>over LAN for instrumentation - it is a part of the LAN extensions for
>>Instrumentation (LXI - http://www.lxistandard.org/), and provides for
>>time sync to tens of nanoseconds (YMMV - routers screw that up). The
>>intended use is in a stack of instruments in a lab, connected to each
>>other and to a PTP source via a local hub.
>>The overall goals and methods of PTP are very much the same as NTP, and
>>I'd hate to have to put both NTPD and PTPD
>>(http://ptpd.sourceforge.net/) on the same machine as they would fight
>>like cats and dogs over who was master - ideally you'd have one daemon
>>doing both, and picking the "best" source.
>>Where I work we are building gear that is Linux based (and so can have
>>NTPD trivially) but I'd like to be fully LXI compliant which means
>>supporting PTP as well (plus some of the stuff we do we need time sync
>>to at worst hundreds of nanoseconds), and being a lazy cuss I'd rather
>>see something like that in the mainstream NTPD rather than having to put
>>it in myself on our own branch.
> Please keep in mind that PTP yields an accuracy in the nanosecond range *only*
> if special network adapters with PTP support are used on all PTP nodes.
> PTP support for network adapters means that the adapters include a hardware
> timestamp unit which takes a time stamp whenever a PTP packet goes on the
> wire or is coming in from the wire.
> Without that hardware support PTP doesn't offer significantly more precision
> than NTP.
More information about the questions