Linda Walsh wrote:
Randall R Schulz wrote:
One my 10.0 system:
% while true; do hwclock; sleep 10; done
Tue 11 Dec 2007 10:25:14 AM PST -0.877900 seconds
Tue 11 Dec 2007 10:25:25 AM PST -0.988284 seconds
Tue 11 Dec 2007 10:25:36 AM PST -0.983046 seconds
On my 10.3 system:
% while true; do hwclock; sleep 10; done
Tue 11 Dec 2007 10:25:16 AM PST -0.000481 seconds
Tue 11 Dec 2007 10:25:26 AM PST -0.000754 seconds
They're running on completely different hardware, but both use NTP.
Randall Schulz
----
Interesting -- using same version of "hwclock" & same
kernel version, I compared 3 machines:
one machine averaged around -0.988xxx, another -0.991xxx
and a third at -0.0003xx. (the xxx digits are variable, the listed
digits were mostly fixed values after the loop start).
Their HW varies considerably, the first two about 5-6 years old,
the latter more recent. I wonder why they cluster like they do -- the
first two around -1, and the newer one nearer 0.
Offset is probably correlated to latency of both
data and interrupts (which get masked in certain
critical sections of code in the kernel).
The newer (faster) hardware has lower latency for both.
--
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
