Section: New Results

Parallelization

Participants : Salvador Abreu, Yves Caniou, Philippe Codognet, Daniel Diaz, Florian Richoux.

During these last decades, many sequential algorithms for Constraint Satisfaction Problems (CSP) have been developed to be able to solve real problems from industry. However these problems become more and more complex and it remains important to treat them as fast as possible. Until the mid-2000's, one developed computers power by increasing CPU frequency. Nevertheless for about five years, this solution is not possible anymore since it asks too much energy (problem linked to heat dissipation issues), thus our machines architecture turns to be more and more multi-core oriented.

Nowadays we still have very few algorithms for constraint problems adapted to multi-core architecture. This year, we obtained very good results with the parallelization of meta-heuristic methods, reaching linear speed-ups over 8,192 cores on the Costas Array Problem [22] , [23] . We also proposed in [20] two ways to perform smart cooperations between parallel local search processes, leading to very promising new approaches to solve constraint-based problems in parallel.