We describe a simulation-based optimization method that allocates additional capacity to transmission lines in order to minimize the expected value of the load shed due to a cascading blackout. Estimation of the load-shed distribution is accomplished via the ORNL-PSerc-Alaska (OPA) simulation model, which solves a sequence of linear programs. Key to achieving an effective algorithm is the use of a High-Throughput Computing environment that allocates computational resources on a platform of more than 14,000 cores simultaneously among several users. We discuss also important implementation details necessary to achieve effective implementation in this massive-scale computing environment. In the end, we demonstrate a prototype computation that reduces the expected load shed by 76% allocating only 1.1% of the installed capacity. The massive-scale computation is made possible using the computational platform provided through HTCondor, effectively obtaining over five months of CPU time in just over one day.