To improve optimizer estimates in case of skewed data distribution , histograms can be created. Prior to 12c, based on No. of distinct values (NDV) in a column two types of histograms could be created :
if no. of buckets >= NDV, frequency histogram is created and the optimizer makes accurate estimates.
If no. of buckets < NDV, height balanced histogram is created and accuracy of optimizer estimates depends on whether a key value is (more...)