Re: [chrony-dev] [GIT] chrony/chrony.git branch, master, updated. 1.25-pre1-18-g20a4340

[Ed W <lists@xxxxxxxxxxxxxx>]

On 13/04/2011 12:10, Miroslav Lichvar wrote:
> On Tue, Apr 12, 2011 at 09:32:31PM +0100, Ed W wrote:
>> On 12/04/2011 15:05, Miroslav Lichvar wrote:
>>> Hm, I just had a crash while I was messing with the tick value outside
>>> chronyd. The sourcestats stddev ended up as -nan which caused assert
>>> failure in find_best_sample_index. It seems to be caused by commit
>>> c88801. The sample weight calculation is not stable when there is a
>>> large jump in distance, the stddev from the regress function is used
>>> in the next weight calculation, which are getting bigger and the
>>> stddev smaller until it reaches the limit of double and fails.
>>>
>>> I'm not sure how to fix it yet.
>>>
>>
>> We live in a finite world?  Bound the result?
> 
> That would fix the crash, but not the underlying cause which I think
> beside bogus values reported in the sourcestats report might have a bad
> effect on frequency stability.
> 
> It seems the problem is that in the weights calculation is used
> weighted variance, which can create the positive feedback. Using
> unweighted variance instead should fix it nicely.
> 

I guess I meant simply sanitise the range of values.

Presumably there is a numerical speed advantage in using weighted
values? Therefore use weighting, but only up to a constrained level.
This will lead to suboptimal convergence in some edge cases, but faster
convergence elsewhere?

Or have a negative "weight" which is some function of volatility. ie
prefer less recent weighting if there is some evidence that more recent
samples are less reliable?

Actually I just re-read your email - perhaps I'm talking cross purposes?
 Sorry..

Ed W

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