Leveraging Enterprise Application Characteristics to Optimize Incremental Aggregate Maintenance in a Columnar In-Memory Database

摘要

An analysis of database workloads generated by enterprise applications revealed a mixed workload of short-running transactional and long-running analytical queries. With the latter type of queries containing many aggregate operations, we implemented an efficient aggregate caching mechanism. But the incremental materialized view maintenance is very costly for aggregate queries joining multiple tables. To overcome this problem, we analyzed the characteristics of enterprise applications with respect to the creation of business objects and their persistence in the database layer. We evaluated how the detected patterns can be leveraged to reduce the join operations between the main and delta partitions of the involved tables in a columnar in-memory database. The resulting performance improvements are significant and close to using the caching mechanism with a denormalized schema.