An Adaptive Aggregate Maintenance Approach for Mixed Workloads in Columnar In-Memory Databases.

The mixed database workloads generated by enterprise applications can be categorized into short-running transactional as well as long-running analytical queries with resource-intensive data aggregations. The introduction of materialized views can accelerate the execution of aggregate queries significantly. However, the overhead of materialized view maintenance has to be taken into account and varies mainly depending on the ratio of queries accessing the materialized view to queries altering the base data, which we define as insert ratio. On the basis of our constructed cost models for the identified materialized view maintenance strategies, we can determine the best performing strategy for the currently monitored workload. While a naive switching approach already improves the performance over staying with a single maintenance strategy, we show that an adaptive aggregate maintenance approach with inclusion of the workload history and switching costs can further improve the overall performance of a mixed workload. This behavior is demonstrated with benchmarks in a columnar in-memory database.