INFORMS Open Forum

Does anyone have real-world experience solving MIPs with Google's CP-SAT and Gurobi?  Any idea when one works better than the other?

I have been asked to provide technical support for what appears to be (I haven't fully dug in) a variation of a set-covering problem that has been formulated as a SAT problem with Google's CP-SAT as the solver.  CP-SAT worked well enough until the problem grew to a few hundred thousand variables and now has to be decomposed and solved in pieces.  There do not appear to be any constraints that can't be phrased as integer constraints and I am convinced that something like Gurobi could easily tackle a problem of that size.

For another well-structured and smaller network flow problem that I have formulated as a straight up MIP using the OR-Tools API, it solves very quickly with CP-SAT although it consistently takes about 3 times as long as Gurobi.  I can deal with it being slower because the solve time is still under a second and the problem is pretty small.  I just don't know how much it will scale.  In that case, though, I can easily switch to Gurobi because I used OR-Tools (whereas CP-SAT problems like the other one can only use the CP-SAT solver). 

As for the first problem, before I go through the tedium of rewriting the problem and the expense of a commercial solver, I'd like to get a little more understanding of where CP-SAT and OR-Tools have an advantage over MIP solvers.