Did you read Carrie Ballinger's blogs on "Statistics Recommendation" for your release? 13/13.10/14?
For some really large tables there's no other way than sample stats.
You just have to be carefull when a column is correlated to a partitioning column, e.g. partition by date and another column year_month. Never collect sample stats on year_month without also including the date, because sampling just starts to read the first blocks and a partitioned table is sorted by partition number...
Dieter
↧