Multi-dimensional data statistics for columnar in-memory databases
Curtis Kroetsch · 2014
The research presented here studies the multi-dimensional data statistics in the context of columnar in-memory database systems. Such systems, for example SAP HANA, Server Apollo, or IBM BLU, use an order-preserving dictionary with dense encoding on the read-optimized storage which encodes the values of a single column in an ordered, dense-domain dictionary. The dictionary maps variable-length domain values to fixed-size dictionary entries. This encoding reduces memory consumption as dictionary values can be represented with fewer bits than the original values, and allows queries to be evaluated efficiently on the encoded data. The main characteristics of the dictionary encoding is that it results in a dense domain of values which can be exploited for building efficient data statistics objects.