Optimization in CUBE computation

• Smallest Parent Computing AB from smallest of its parents ABC and ABD is more efficient.
• Cache Result Compute AB while ABC is still in memory.
• Amortize Scans Compute ABC, ACD, ABD and BCD in one scan of ABCD.
• Share sorts Share sorting cost across multiple group-bys in the sort based algorithms.
• Share partition In hash based algorithm, when hash table is too large, it must be partitioned. This partitioning cost can be shared across multiple group bys.
• Overlap optimization 1 An aggregate can be computed from a sorted ascendent aggregate in sorted order.
• Overlap optimization 2 An existing sort order on an aggregate can be used while computing others from it. For example, consider ABD to be computed in the same sorted order from ABCD, which is sorted in the same order. In that case, one need not resort ABCD to compute ABD but use the existing sort order of ABCD.