Re: NTP accuracy?

H. Peter Anvin (hpa@transmeta.com)
2 Dec 1998 05:35:15 GMT


Followup to: <199812012105.NAA10340@bitmover.com>
By author: lm@bitmover.com (Larry McVoy)
In newsgroup: linux.dev.kernel
>
> hi folks,
> excuse the ignorance of this question, but suppose you had 1,250
> systems hooked up to a networking fabric with 10usec end-end propogation
> times, but the times can vary depending upon contention. What I want to
> know is this: how accurately can NTP sync the clocks across all those
> machines? I know it may be asking for too much, but is it possible to
> get 1usec accuracy? If it makes it any easier, this is a controlled
> environment, so it is possible to have a tree structure in the nodes
> such that the times fanout very quickly to all nodes.
>
> If it can't be done with NTP, anyone have any suggestions on how to do this
> with a software solution?
>

No, you probably cannot get ▒1 Ás with any kind of software solution
in this environment. The dispersion is simply going to kill you ("the
times can vary depending upon contention" is the deadly phrase.) NTP
is very good, but there are things it just can't do. I suspect you
will be able to get about ▒200 Ás with NTP in this environment without
trying, and can probably shake it down to ▒50 Ás with a well-designed
tree.

Incidentally, you probably want your primary source of time to be at
least an order of magnitude more precise than you want your clients.
▒100 ns GPS clocks are standard issue these days, so it shouldn't be a
problem.

-hpa

-- 
    PGP: 2047/2A960705 BA 03 D3 2C 14 A8 A8 BD  1E DF FE 69 EE 35 BD 74
    See http://www.zytor.com/~hpa/ for web page and full PGP public key
        I am Bahß'Ý -- ask me about it or see http://www.bahai.org/
   "To love another person is to see the face of God." -- Les MisÚrables

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/