RAM: array processing over a relational DBMS
Developing multimedia applications in relational databases is hindered by a mismatch in computational frameworks. Efficient manipulation of multimedia data calls for array-based processing, which at best is available as a database add-on, not supported by the query optimizer. As a result, array-based processing ends up in dedicated programs outside the DBMS: non-reusable black boxes. The goal of our research is to reduce this gap between user-needs and system functionality by developing a seemless integration of array processing in a relational algebra engine. The paper introduces a declarative language for array-expressions based on the array comprehension, and its mapping to a relational kernel in a prototype implementation. The layered architecture of the resulting array database management system allows the use of structural knowledge available in the array data type. This additional source of information can be exploited for query optimization, which is demonstrated with a case study. The experiments show how the performance of a standard tool for matrix computations can be achieved without sacrificing data independence, highlighting however a critical aspect in the DBMS architecture proposed.