Constrained Optimization Over Massive Databases

Constrained optimisation is increasingly considered by industry as amajor candidate to cope with hard problems of practical relevance.However, the approach of exploiting, in those scenarios, current Con-straint or Mathematical Programming solvers has severe limitations,which clearly demand new methods: data is usually stored in poten-tially very-large databases, and building a problem model in centralmemory suitable for current solvers could be very challenging or impos-sible. In this paper, we extend the approach followed in [3], by present-ing a declarative language for constrained optimisation based on sql,and novel techniques for local-search algorithms explicitly designed tohandle massive data-sets. We also discuss and experiment with a solverimplementation that, working on top of any DBMS, exploits such al-gorithms in a way transparent to the user, allowing smooth integrationof constrained optimisation into industrial environments.