Real Life Performance of In-Memory Database Systems for BI
by Dr. Andreas Hauenstein, Dr. Simon Hefti and Dr. Andrej Vckovski
Recently, quite a few main memory database systems have been entering the
market. They rely on cheap 64-Bit hardware with huge amounts of RAM where
traditional database systems tend to optimize disk access. Performance
problems with a fairly standard data warehouse design at a customer
prompted us to evaluate a few of these new systems by evaluating their
performance on our real life problem.
A synthetic performance benchmark framework was developed, which mimicks the customer data model. It allows for reproduction of the observed performance problems on any system without much effort. The database systems involved were Oracle, SQL Server, MySQL, LucidDB, Exasol, and MonetDB. While the first three are well known, mature systems, the other three are much younger, column-store based systems. We present the framework used for testing, discuss the tested systems, and report the experiments and their results.
» Program of TDWI 2010 Munich, Germany
» Presentation T1P, June 15,2010, 2:45 to 6:00pm at TDWI 2010