> The elevator pitch behind this performance difference is > straightforward: column-stores are more I/O efficient for read-only > queries since they only have to read from disk (or from memory) > those attributes accessed by a query.
> This simplistic view leads to the assumption that one can ob- > tain the performance benefits of a column-store using a row-store: > either by vertically partitioning the schema, or by indexing every column so that columns can be accessed independently. In this pa- > per, we demonstrate that this assumption is false.
https://faculty.cc.gatech.edu/~jarulraj/courses/4420-s19/pap...
