Field specialization is Dataware's new technology for dynamic code optimization. The primary objective of field specialization is to increase the performance of an application without requiring any major changes to that application's source code.
Field specialization eliminates overheads arising from overly-general code that is unnecessary for particular inputs, resulting in code that is significantly smaller and faster.
Field specialization has been applied and validated on multiple code bases, including four proprietary and four open-source DBMSes. This demonstrates that our technology is effective on code bases comprised of millions of lines of code developed over decades and is broadly applicable.
Just a few specializations produce significant speedups across workloads. The specializations are small, each ranging from two dozen to two hundred lines of code, and thus easy to add and maintain compared to other approaches. Specialization does not change the semantics of the DBMS. Specialization is applicable at scale, such as in the cloud, as each specialization can benefit a broad range of workloads. The speedups we obtained on the eight DBMS products were significant in all cases.
Field specialization can deliver DBMS performance improvements more rapidly than conventional approaches. To add and test a particular specialization (starting from no experience with the source code of each DBMS) required about a person-month on average.
We have a proprietary development process and associated tool set to facilitate that process. This process and tools help us speed up individual routines, which in turn open up additional specialization opportunities in other routines.
Further Information White Paper