Towards Generating Application-Specific Data Management Systems

Alvin Cheung

Abstract for CIDR 2015


Specialization is an effective way to improve performance. That principle has manifested time and again in database management systems (DBMSs), with custom data storage engines such as column stores and streaming engines outperforming traditional DBMSs in specific application and architecture domains. Unfortunately, designing custom data storage engines is a tedious task. Developers need to first analyze numerous applications and their hosting platforms to identify common patterns in manipulating persistent data. This is followed by a painstaking and trial-and-error process of devising efficient implementations for the data operations on the target platform. We argue that advances in program analysis and software synthesis can help in automating this design process. In the following we describe opportunities where understanding the application can improve functionalities provided by the DBMS.


[Full Abstract]