Can field specialization help your DBMS?
Dataware speeds up relational databases through a new technology we call field specialization. Just how well might field specialization work with your DBMS?
To help answer this question, we applied Dataware’s field specialization to three DBMSes and evaluated its effectiveness using the TPC benchmarks.
We applied a handful of specializations to the 380,000-line PostgreSQL DBMS, touching less than 1% of the code, but producing big improvements in execution time: speedups of up to 3X for the top queries and at least 1.7X for all queries.
We also applied one field specialization to MonetDB, a modern columnar-store architecture with query compilation and one field specialization to VoltDB, a recent extremely high-performance DBMS targeting OLTP applications. Again, the actual code changes were small. When we ran the specialized DBMSes on a single query from the TPC-H and TPC-C benchmarks respectively, in each case, field specialization achieved improvements of over 20%.
The Postgres experiments show that each field specialization improves performance and that, together, many field specializations can significantly reduce DBMS run times across complex analytic queries, random modifications, and bulk-loading. The MonetDB and VoltDB experiments show that field specialization can be effective across these very-different DBMS architectures.