They introduced a technique called 'vectorized execution' to improve the performance of database queries. In vectorized execution, all database operators are not executed one by one on individual database records, as in previous systems, but always on small columnar blocks called 'vectors'. As a result, analytical systems are at least ten times faster. "The principle is relatively simple," says Peter Boncz. "It keeps the database software easy to maintain. It also helps with performance profiling, which is important for optimizing slow queries."
The winning paper was part of Marcin Zukowski's PhD research at CWI. Together with Boncz and CWI researcher Niels Nes he created the VectorWise database system and CWI spin-off in 2008, the fastest analytical system at the time.
Zukowski went on to co-found Snowflake in 2013, which is now one of the largest cloud database services. And not only Snowflake uses vectorized execution, but Databricks, Google BigQuery, Microsoft's SQLserver and many other systems have since adopted the technique. It is also used in CWI's new database system, DuckDB, and the MotherDuck cloud service, which uses DuckDB.
The CIDR organization selected the vectorized execution paper from 52 papers presented at their 2003 and 2005 conferences. The paper, 'MonetDB/X100: Hyper-Pipelining Query Execution', is the first to receive the Test of Time Award. The award was presented Monday 15 January at CIDR 2024.
Lecture
Afterwards, Peter Boncz and Marcin Zukowski gave a lecture on the development of the VectorWise database system and the legacy of the winning paper.