Daniel Carosone <dan@geek.com.au> writes:

> I recall some discussion in the xen documentation about a xen clock
> device intended to expose to each domain both the real and the
> domain-virtual elapsed time.  Are we using this, and are we using it
> correctly?  I would have hoped that this would render ntpd's in domU's
> unnecessary.

But for real time, the question remains whether it is more-or-less
real as measured by clock interrupts in dom0 (and therefore has an
initial phase error and a frequency error) or Xen itself (same issues
I would think), or real time as adjusted by ntpd in dom0.

My experience is that domUs without ntpd end up a few seconds off,
which supports the theory that basic timekeeping is ok, but that
frequency error is not corrected.

-- 
        Greg Troxel <gdt@ir.bbn.com>