Building OLAP Data Analytics by Storing Path-Enumeration Keys into Sorted Sets of Key-Value Store Databases

Abstract—In this paper, a novel approach that allows the retrieval of OLAP analytics by means of storing multidimen-sional data in key-value databases is described and evaluated. The proposed mechanism relies on generated keys based on the path-enumeration model built upon a tree representation of a relational database. Each key represents a combination of OLAP dimension labels and is inserted into a key-value sorted set where its associated values are the metrics of interest, i.e., the OLAP facts. Our implementation for a real advertisement server system in Redis and its performance comparison with a relational OLAP (ROLAP) database running on top of MySQL, show that our proposed scheme can answer complex multidimensional queries much faster, while keeping a great flexibility of the data schema. In contrast with general ROLAP schemes, which usually require the pre-computation of tables for specific queries, the sorted sets in the proposed system are not pre-computed but generated on demand, so no further delay is introduced. To the best of the authors ’ knowledge, this is the first system that deals with mapping relational data structures into sorted sets of key-value store databases to answer complex queries on multidimensional data. Keywords-OLAP; path-enumeration keys; sorted sets; databases; key-value stores. I