Improving tuple reconstruction for tiered column stores
Martin Boissier, Alexander Spivak, Carsten Meyer · Proceedings of the Australasian Computer Science Week Multiconference · 2016
Improving memory consumption of main memory-resident databases by evicting infrequently accessed data to secondary storage layers can significantly reduce the TCO. But traditional approaches for data eviction such as paged-based LRU algorithms are unsuitable for OLxP systems with their scan-dominated workloads. In this paper, we present a data tiering ansatz for columnar in-memory databases that is workload-aware and retains full performance for analytical queries. To make our tiering ansatz also applicable for transactional workloads, we will focus on the topic of table reordering to improve tuple reconstruction, which is one of the the main bottlenecks for tiered columnar databases. We evaluate our approach using data and workload of a production SAP ERP enterprise system of a Global 2000 company.