Wed Oct 18 20:20:18 NZDT 2006
On 10/18/06, Glenn Ramsey <glenn@c...> wrote:
> Hi Ian,
> Ian McDonald wrote:
> > I'm wanting to sync two machines to within 1 msec for my research using ntp.
> > Questions:
> > - is this realistic given how clocks drift? Has anybody done this?
> > - is there a quick way to force them to converge
> > I've had a quick read of the ntp pages and made one of my machines a
> > broadcast server and the other broadcast clients I think but they seem
> > to be taking a while to converge. The man pages say they will fix at
> > 0.5ms/sec but I'm not seeing that.
> > Comments appreciated.
> > NB I don't want my clocks to be deadly accurate to the "real world",
> > just to each other.
> A while ago I looked into this a little and from what I recall 1ms is a
> big ask for a commodity PC, 10ms I think was achievable. As I recall it
> a major issue is the accuracy of the PC's internal clock. NTP does
> account for this by working out how much the PC's clock drifts but that
> still only gets it so far.
> I think to get that accuracy you need to equip your PC with a more
> accurate local clock. I think some sort of GPS card is commonly used.
> There's screeds of info about this issue on the NTP related websites.
> Google should find it.
> Also there's another protocol around that is used for syncing PC's but
> can't remember what it was called.
> That probably wasn't that much help, but it might help you to look in
> the right places.
Thanks to everybody's comments and feedback.
I've actually managed to code up something myself by building a
simplified NTP and it gets me down to around 1 msec.
WAND Network Research Group
Department of Computer Science
University of Waikato
More information about the wlug
NOTICE: This is an archive of a public mailing list. The University of Waikato is not responsible for its contents.