[ntp:questions] Severe time drift for Undisciplined Local Clock

David Woolley david at djwhome.demon.co.uk
Wed Apr 11 21:24:58 UTC 2007


In article <20070410201736.t1uxq69cu8cgso4k at webmail.nulan.co.za>,
deon at nulan.co.za wrote:

> the collective wisdom and have located a  Windows XP machine at each  

I don't think we would ever advise using any Windows machine in this
context and we certainly wouldn't suggest one running W32time, which
this one appears to be doing.

What we might advise is a 16MB 486 DX running FreeBSD (or Linux).  The most
important thing, though, is to run the reference implementation of 
ntpd.  (Linux is OK for a dedicated server, although it can lose clock
ticks if it is doing other things that require disk I/O.)

> rootdispersion=10414.490, refid=LOCL, reach=377, unreach=0, hmode=3,

One of the characteristics of the most recent version of W32Time (older
ones are worse) is that it reports a realistic rootdispersion value for
undisciplined local clocks, i.e. it is quite reasonable that the clock
may be 10 seconds off true time.  That means that you cannot use it as
a local clock source of bogus time for compliant clients.  The reference
implementation assumes that the local clock is being disciplined, but
by something outside its control, so reports a low root dispersion.

If you are stuck with Windows, install the reference ntpd and disable
W32Time.  I think it supports the local clock driver.

PS Don't start subjects with Re: when you are starting new threads.
Do include at least In-Reply-To (for the mail gateway), if continuing
a thread. 




More information about the questions mailing list