Fast computation of Iceberg Dwarf

Iceberg Dwarf (IceDwarf for short) combines the strength of Iceberg-Cube and Dwarf. It exploits the elegant Dwarf structure for cube tuple store and eliminates those unsatisfied sub-dwarfs. By only storing nontrivial cube tuples, IceDwarf reduces the size of a dwarf significantly; even Dwarf itself compresses the data cube effectively. We studied how to efficiently compute icedwarfs, and developed a straightforward algorithm (PAC). To further improve the performance, we explored the structure of Dwarf and presented four nice lemmas. Based on these observations, we proposed a new algorithm called PWC. It builds the IceDwarf by bottom-up computing all the partitions of a fact table and inserting them into the Dwarf structure, enabling Apriori-like pruning and single tuple partition optimization, and facilitating the detection of suffix redundancies. Our performance study showed that PWC is highly efficient and runs much faster than PAC for icedwarfs, even for computing full dwarfs.