Sandia National Laboratories has invested considerable effort in massively parallel optimization tools. In this talk, we will summarize relevant experience from developing our parallel mixed-integer programming (MIP) solver PICO (Parallel Integer and Combinatorial Optimizer) and our parallel nonlinear solver GNLP. We will discuss parallel solution of mixed-integer nonlinear programs (MINLP). In particular, we will consider how the choice of parallel platform (tightly-coupled systems, cloud computing, grid computing, etc) can affect algorithmic decisions. We will highlight issues parallel solvers must face that serial solvers do not such as load balancing, ramp up, and termination and some issues that parallel solvers might do differently, such as decomposition. We expect some of our MIP/NLP experience to carry over, and some issues to be unique to, or uniquely difficult for, MINLP.