Solving several model instances in parallel

In this section we show how to execute several models in parallel and communicate solution information among these models. This scheme may be particularly interesting when working with Mosel on a multi-processor machine, e.g. by starting a number of models that corresponds to the available number of processors.

Our idea is to run several instances (different only by the parameterization of the solution algorithm) of the same MIP model concurrently and to stop the entire run when the first model has finished. If the different solution algorithms are complementary in the sense that some quickly produce (good) solutions and others are better at proving optimality once the best solution is found then one may reasonably expect an additional synergy effect from exchanging solution updates during the MIP search.

To implement this scheme, we define a master model that starts the model runs and coordinates the solution updates, and a parameterizable child model that is loaded and run with the desired number of versions. The child models all use the same solver (Xpress Optimizer) but it would equally be possible to use a different solver for some of the child models, provided it defines the necessary functionality for interacting with the search.