On Fri, 2007-08-24 at 21:53 -0400, Brian Utterback wrote:
> Well, nothing can make a system clock is not well behaved do
> very well. NTP can only correct up to an error of 500 parts
> per million, which is still pretty small.
For a clock, 500ppm is a huge error. That's equivalent to an error
(loss or gain) of 43 to 44 seconds per day.
Mechanical clocks have been able to do much better than that for
centuries. One notable 18th century clock (John Harrison's "H5")
is reported to have had an error of no more than 5 seconds in 10 days --
or about 5ppm.
Running inside vmware no doubt makes things a bit more difficult -- I'd
think that if our timekeeping inside vmware sucks it's a bug (perhaps a
poor interaction between what vmware provides and what we expect for
timekeeping). Rather than expect NTP to paper over the errors it might
be better to talk to them about the problem and see if they have any
suggestions...
- Bill
