In this paper, we develop the first theoretical formulation of the
power-constrained performance optimization problem for MPI +
OpenMP applications. We present an optimal verified linear pro-
gramming (LP) formulation, which places a theoretical upper bound
on the application performance that can be achieved on power-
constrained systems. The LP formulation is optimal under certain
conditions. We also present an integer linear programming (ILP)
formulation of the same problem, which, while optimal, is not fea-
sible for realistic problem sizes. However, we show that the LP and
ILP formulations yield similar results, and that the LP allows us to
find theoretical limits for realistic problem sizes.