Re: [chrony-users] Isolated time domains

[ Thread Index | Date Index | More Archives ]

On Thu, 5 Dec 2013, Miroslav Lichvar wrote:

On Wed, Dec 04, 2013 at 12:46:10PM -0800, Bill Unruh wrote:
On Wed, 4 Dec 2013, Miroslav Lichvar wrote:
With the cosine smear, the frequency offset changes slowly (small
wander), but can reach a large absolute value.

If I calculate right, it can corrrect an offset of 10000 seconds over
1e6 seconds (11.6 days) with maximum wander of 0.05 ppm/s and maximum
frequency offset of 15915 ppm.

Althought I'm not sure if the cosine function is the best choice here
as the wander is at the maximum at the start (2nd derivative) and the
clients are possibly using their longest polling interval and will
react slowly.

I guess the above was what I was worried about. The clients take a long time
both the poll times, and the reaction time-- the time to realise that the
offsets imply that their own frequency has to change, which is many poll
intervals. Ie, even at the minimum of poll 6 (a minute) it will take many
minutes to react to a change in frequency in the client, and an attempt to
close the offset gap. And the huge change in frequency makes (15000PPM)
makes it even harder since chrony tends to resist large changes. Ie,
I would expect the clients to end up all over hell's half acre in

But the frequency doesn't change instantly by 15000 ppm, it takes half
of the 11.6 day period to reach that value and the other half to get
back to the original frequency. At the beginning it changes by 0.05
ppm per second, that's about 50 ppm in 1024 seconds.

You mean a sin funtion. But for such large changes in rate, I would still worry about the clients
ability to follow.

In this example I think clients running chronyd would stay together
within 1 second during whole operation even with 1024s polling

That is what I am not sure of.

For situations like this it might be a point to have chrony not just fit a
straight line to the data, but a quadratic. That would also fit the
acceleration rather than just the drift, and would allow it to keep up. Since
one of the primary places where time gets difficult is if the temperature
changes, and the temp change changes the drift rate (acceleration) it might
also allow chrony to track temp changes better as well. But it is not
completely clear that the increase in complexity would bring a concomitant
increase in tracking ( or that doing that might not introduce instabilities as

Anyway my feeling is that trying to do what he says he wants if worth the
increase in complexity. The best way to keep a set of clocks synchronized is
to synchronize them all to UTC. If one of the clocks (eg one of the clients
or the server)) goes bonkers for a while, I think letting it find its own way
home is better than all of the computers then trying to chase it.

Eg, he says that he wants to keep all of the computers to within 1 sec. But
what happens if one of the clients suddenlyfinds itself 3 hours out? Are all
of the other clients then supposed to go out as well to keep themselves in
sync with that client? How would they do that? And if not, why is the server
special that all should run after its crazyness?

The initial response could be improved if the wander was a sine
function or triangle instead of cosine. The frequency would start
changing slowly and the clients would have time to shorten their
polling intervals and follow closely faster frequency changes later.

It would still be a poll interval of 6 with the default (that is a minute) and
chrony needs at least 3 polls to fit a straight line to, so that is 3 min, and
to get to 15000PPM in 5 days, say, chrony would still have to change the drift
by 2 PPM/poll (or 6PPM per measurement) which would accumulate over the 5 days
to about a few  seconds. and they would still be way off in time since the
master IS out by 3 hours, which each of the clients think they have to fix.

William G. Unruh   |  Canadian Institute for|     Tel: +1(604)822-3273
Physics&Astronomy  |     Advanced Research  |     Fax: +1(604)822-5324
UBC, Vancouver,BC  |   Program in Cosmology |     unruh@xxxxxxxxxxxxxx
Canada V6T 1Z1     |      and Gravity       |

To unsubscribe email chrony-users-request@xxxxxxxxxxxxxxxxxxxx with "unsubscribe" in the subject. For help email chrony-users-request@xxxxxxxxxxxxxxxxxxxx with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.

Mail converted by MHonArc 2.6.19+