On Sun, Jun 28, 2009 at 01:04:51PM +0200, Robert Olsson wrote:
...

 Yes we keep the old tnode size and the convergence interval was some
 of the concerns. That why this checks was added. Still we want to
 inflate the root node to a very max. 

 So this approach with halving or doubling tnodes towards the root
 node was the suggest by Dyntree paper. I asked Stefan (one of the
 authors) if we could get safe and very offensive settings. But 
 from what I understood there was no easy way to calculate this. 
 So any bright ideas in this area are very welcome. But we should
 also monitor size of root and average tree depth so we don't
 take an to defensive approach just to solve this case.

Yes, but with this offensive approach it seems the current level of
warnings could be too alarming. Btw., because of a design flaw in my
current patch this _fix variable, which should be logically per trie/
table, can be reset by changes of other tables now, but I think it
all could be fine tuned more in the future. Of course if there are
people interested in testing/reporting this more.

Thanks,
Jarek P.