I am currently developing a column generation algorithm for path-based network flow problem. According to the Gurobi website, by default Gurobi uses the concurrent optimizer for LP that runs multiple solvers (Dual SM, Primal SM, and Barrier) on multiple threads simultaneously and choose the one that finishes first. But I noticed that in my CG problem, it was even faster when I forced Gurobi to use Barrier method. When I run Gurobi with the default algorithm, it chooses Barrier in the 1st iteration and SM for the rest of the iterations. According the description of the Gurobi method, the default algorithm chose SM for the rest of the iterations since it found SM faster right? but Barrier is performing a lot faster... It seems like something is forcing Gurobi to use SM... Does anyone have a clue why it is happening?
Please sign in to leave a comment.