Equipment failures, operator errors, natural disasters and cyber-attacks can and have caused extended blackouts of the electric grid. Even though such events are rare, preparedness for them is critical because extended power outages endanger human lives, compromise national security, or result in economic losses of billions of dollars. Since most of the generating units cannot restart without connecting to an energized grid, the system operator relies on a few units with the ability to start autonomously, called Black Start (BS) units, to restore the power system. Allocating and maintaining these units is costly and can severely impact the restoration security and time. We formulate an optimization problem to optimally allocate BS units in the grid, while simultaneously optimizing over the restoration sequence. We extend existing optimal allocation models by including grid considerations such as active power nodal balance, transmission switching, nodal reactive power support and voltage limits. In order to aid the branch and bound tree that solves the resulting large scale Mixed Integer Program (MIP), we propose a randomized heuristic that is executed multiple times in parallel on a high-performance computing environment to find feasible solutions. We proceed to solve the IEEE-39 and the IEEE-118 systems to near optimality.