Dynamic partitioning for enterprise applications

Today´s enterprise applications face a severe change in how they process data. They evolved from simple data entry applications to complex systems where the focus is to make the right decision based on latest transactional data instead of pre-processed outdated business intelligence data from secondary systems. This paradigm shift results in more analytical queries executed on the transactional database system. In addition we see a change in how the persistence layer is seen and that more and more effort is spent on developing database engines that are designed to address special needs and requirements, e.g. for Web 2.0 applications. In this paper we discuss that it is crucial for enterprise applications to share application semantics with the database to increase performance. Furthermore we show how this principle can be applied to dynamically partition application data. In contrast to static partitioning we propose to adapt the life-cycle semantics of the application and move data from active to different passive partitions, by leveraging the merge process of our database prototype. The goal is to reduce the amount of data that is touched during query execution to increase the performance of the main workload by the same factor. We present how our HYRISE prototype is implemented and how it can be extended to fully support this use case.