Thanks Alberto! It was w good read (11g new density). One question: You have mentioned that CBO classifies at run-time the histogram type as it's stored in DD using a heuristic. As per my knowledge, based on NDV and #Buckets requested, CBO decides whether to create HB or Freq histogram and save the type in data dictionary, which is visible through dba_tab_columns.histogram column. Please correct me! On Thu, Apr 9, 2009 at 3:44 AM, Alberto Dell'Era <alberto.dellera@xxxxxxxxx>wrote: > On Mon, Apr 6, 2009 at 17:20, Neeraj Bhatia <neeraj.dba@xxxxxxxxx> wrote: > > Basically I am interested to know how density is calculated in case of > > height-based histograms. Also, how CBO calculate selectivity. > > I wouldn't investigate too much the density in 10.x since the formula > has changed (to a very intuitive one) in 11gR1: > > http://www.adellera.it/investigations/11g_newdensity/index.html > > so, the 10.x formula has been "deprecated" by the kernel developers. > > I did find a statistical justification for the 10.x computation, but > I have never "published" it for the above reasons, it is no longer > important. > > hth > alberto >