Guillaume Derval graduated with Engineering and Computer Science master from UCLouvain in June 2016, and since then works as a Phd student on the subject from his master’s thesis: the usage of concurrency, parallelism and distribution in Constraint Programming.
He is also one of two main developers of INGInious (http://www.inginious.org), a secure and highly customizable computer science exercise grading tool, used by most CS courses at UCLouvain/INGI and also in other universities around the globe.
OscaR-Modeling is a new modeling layer for OscaR, a Scala toolkit for solving Operations Research problems.
The modeling layer abstracts the model (the problem you want to solve) from the solver (the way you want to solve it), leading to new possibilities with concurrency, parallelism and distribution, which is in fact the primary goal of this new implementation. The new layer adds the concept of model tree and model concretization to OscaR, making them completely declarative and allowing to reuse a given model efficiently on multiple different solvers. As a bonus, OscaR-Modeling opens new possibilities about pre-solving in OscaR.
OscaR-Modeling contains the first efficient implementation of Embarrassingly Parallel Search (EPS), the new state- of-the-art for parallelizing and/or distributing Constraint Programming computations. Tests on our implementation have shown perfectly linear speedups on the problems tested, even on very large computing grids, with for example a speedup of ≃ 980 on a 1024-core cluster for a Golomb-Ruler.
The software was developed during Guillaume master’s thesis about EPS, and the work continues with the start of a Phd thesis with the same subject.