List comprehensions in Agna, a parallel persistent object system

Abstract

List comprehensions are structurally similar to SQL, the standard declarative query language for relational databases. Unlike SQL, they are more general, and are embedded seamlessly into powerful programming languages. Thus, functional languages with list comprehensions are attractive as query languages for persistent object systems, where database objects and queries are not distinguished from other objects and computations. We have implemented such a system, called Agna. In this paper, we describe optimizations on list comprehensions in Agna. We use techniques borrowed from relational databases, as well as techniques specific to our parallel implementation. Experimental results show a dramatic impact on performance. A comparison with a commercial relational database system and with an experimental parallel relational system indicates that the performance of Agna approaches that of state of the art relational database systems.