Sampled Stats vs Time trade-off - response (3) by dnoeth

For a table with a PI but no partitioning the sampling approach is usually ok, a table is sorted by RowHash -> not based on the order of inserts and looks quite random (for a NoPI table it *is* a chronological order).

When you SAMPLE on a partitioning column in a PPI table the optimizer is smart enough to know about that and samples the first x percent from *each* partition.

But there's not way to change the optimizer strategy (although it would be nice to have regarding the problem with correlated date)

Dieter