Sometimes I see solver logs that makes me wonder… Here is an example:
MIQP Presolve eliminated 1 rows and 1 columns.
I.e. 5 seconds to solve the relaxed qp and then spending 775 seconds in the crossover (without a good result)!
It is good to see that there is progress in this area (quadratic related optimization). E.g. from a recent presentation:
(http://www.slideshare.net/IBMOptimization/euro-2015-nodet). In general my feeling is that quadratic solvers are not as mature as linear solvers, and occasional show erratic behavior. As I see typically difficult models (for easy ones users don’t need my help), I encounter issues like these more than I care for.