[time] Accuracy goal of the pool?

Michael Zimmermann zim
Sun Nov 19 15:03:26 UTC 2006


Hi Rob, hi list,

Rob Janssen schrieb:
> ... when you take a good ntp daemon and let it
> sync to a server over the Internet, it certainly should be possible
> to be within 10ms.
Sure. But if you run the ntp daemon on a non-dedicated work-horse
machine with normal server hardware (as I do) the clock may jump away
from time to time for more than 10 msec - due to sporadic load
of the machine or the like.

See the results of one of my servers as seen by the pool.ntp.org
measurement:

http://www.pool.ntp.org/scores/83.133.111.7

and compare it with the offset of that same machine as measured
from other time servers in my net:

http://k4.khidr.net/tmp/tictoc.php

the last three images of that page (after the heading
"Offset of reference server - clock on k4") show offset, delay
and jitter of the nearest stratum 1 server as seem by the NTP
daemon on 83.133.111.7.


Monday evening for example there was a jump of about 140 msec
in offset. Monday evening is famous to have extremely heavy load
on the webserver of that machine.

While during "normal" times, now and then an offset of up to 20 msec is
no uncommon.

The effects of bad network conditions can be seen from Friday evening
thru Saturday.


Hence 100 msec  max offset seems to be a reasonable limit for
ntp-daemons running on "normal" modern internet-servers
(the "k4" hardware uses a Tyan mainboard by the way). Offsets below
10 msec can nearly always be seen in the peers-query, but
the more seldom "breakouts" are sometimes above 10 msec.


Michael

-------------- next part --------------
A non-text attachment was scrubbed...
Name: zim.vcf
Type: text/x-vcard
Size: 331 bytes
Desc: not available
Url : http://fortytwo.ch/mailman/pipermail/timekeepers/attachments/20061119/247203af/attachment.vcf 



More information about the pool mailing list