Please find below some advise:
(1) Avoid choosing Boolean, Floatiing point datatypes as distribution columns. Boolean causes data skew. Floating column cause Merge sort join.
(2) Pick up commonly joined column(s) as distribution column
(3) Note that you can’t update distribution column
(4) If possible, avoid multi-column distribution columns. Go for RANDOM in such scenario!
(5) Its advantageous not to worry so much about dimensional tables distribution key. Prefer choosing RANDOM for dimensional tables
(6) Use same data types for distribution between joined columns
(7) Don’t distribute on Random because it is the easy choice