On Tue, Dec 06, 2016 at 11:17:26AM -0800, Denny Page wrote:
> Yes, the general server that I refer to, and which the graphs are from, is 
> fairly active on the primary nic. It’s the network log server, does some DB, 
> etc.
> 
> On the quiet system, the change has much less impact. But it was the general 
> server that I was frequently seeing small spikes on.
> 
> I can certainly load the network a lot more if you would find it beneficial 
> for me to test that.

Yes, I think that might be helpful. I spent some time today comparing
your method with the current code and at least on my system with i210
I see a shift in the distribution of the offset to one side when the
network is (heavily) loaded. Compare these two histograms

http://i.imgur.com/GsRwhyX.png (min delay * 1.1)
http://i.imgur.com/MHYRCCx.png (min delay + sys_prec)

Here is a scattergram of the offset vs delay, ignoring the first
reading. It shows there is a strong asymmetry.

http://i.imgur.com/EgPL9mH.png

At least for now, I think I'll leave the code as it is. We can improve
it later when we have more data. There are other things that may need
to be implemented in order to improve accuracy with HW timestamping,
like dynamic PHC reading interval for instance. There is certainly
a room for further improvements.

I have a couple of unrelated changes in my queue and then I'll spend a
day testing the new features, so I can finally make a 3.0 prerelease
and more people can test it. If anyone sees any bugs, please let me
know.

-- 
Miroslav Lichvar

-- 
To unsubscribe email chrony-dev-requ...@chrony.tuxfamily.org with "unsubscribe" 
in the subject.
For help email chrony-dev-requ...@chrony.tuxfamily.org with "help" in the 
subject.
Trouble?  Email listmas...@chrony.tuxfamily.org.

Reply via email to