At the research group of Prof. Plattner, one of our goals is to further improve in-memory databases by developing better algorithms and leveraging new hardware technologies. We implement these new concepts in our research database HYRISE, designed and developed at the Hasso Plattner Institute.
HYRISE is an open-source in-memory database, available for anyone interested at github.com/hyrise/hyrise. It is based on a primarily column-based data layout, allows for data compression and handles updates by storing them in periodically compressed data chunks. Its lightweight setup makes it easy for researchers to change components and compare them to existing approaches. In the past year, we have fundamentally rewritten HYRISE towards a better NUMA awareness, beginner-friendly code, and higher efficiency.
HYRISE leverages an upcoming type of memory, so-called Non-Volatile RAM. By storing data structures directly on NVRAM, HYRISE allows for instant restarts of the database without any recovery times. HYRISE also aims at replicating databases to satisfy the growing analytical processing demand of enterprise applications. A dedicated master instance does all the transactional workload. Read-only replica instances cope with the analytical queries. This allows an elastic scale-out with optimized replica configurations specialized for analytical queries.