High-Performance Data Management Beyond DRAM
Victor Leis, TU Munich
Abstract: In the past decade, main-memory database systems like SAP HANA, Microsoft Hekaton, and Tableau Hyper have gained widespread adoption. These in-memory database systems offer very high performance, but have one major weakness: They expect all data to reside in DRAM. In this talk, I will first present LeanStore, a novel storage engine for modern hardware (fast SSDs, large DRAM capacities, many CPU cores). LeanStore is almost as fast as in-memory systems as long as the working set fits into main memory, but gracefully and transparently handles very large, SSD-resident data sets as well. In the second part of the talk I will discuss how to best utilize the upcoming byte-addressable non-volatile memory (e.g, 3D XPoint DIMMs) in database systems. I argue that this technology will replace neither DRAM nor flash, and is therefore best used as an additional layer in the storage hierarchy that is automatically managed by the database system.
About the speaker
Viktor Leis is a postdoctoral researcher at the Technical University of Munich, where he earned his doctorate in Computer Science in 2016. Viktor does research on core database topics, including query processing on modern hardware, query optimization, concurrency control, index structures, and storage. Much of this research has been integrated into the HyPer main- memory database system, which was acquired by Tableau in 2016. He received the biennial dissertation award of the German-speaking database community (GI DBIS) 2015/2016 and a best paper award at ICDE 2014 for his work on hardware transactional memory.
Date & Time
Monday, April 9, 2018 - 14:00
Huxley 218