Application-aware aggregations in a columnar in-memory database

The reunification of online transactional processing and online analytical processing in one single database has been proposed recently as the advent of columnar in-memory databases makes this both viable and feasible. Compared to traditional row-based databases, resource-intensive operations such as the aggregation can now be executed with high performance. This has emerged to the radical approach of eliminating materialized aggregate tables and calculate all aggregations on the fly. This paper contributes by investigating application and data characteristics that have an influence on the aggregation performance in columnar in-memory databases. In particular, we have examined the Availability-to-Promise application that calculates if a desired order can be fulfilled on the requested date and in the requested quantity. We have implemented a prototype that incorporates our identified characteristics and backed up our findings with relevant experiments and real-life customer data.