Accelerating a Many-Scenario Water Resources Simulation

We present a new approach to water resource simulation that exploits modern CPU hardware and instructions. The approach simulates multiple scenarios in parallel using single-instruction, multiple-data (SIMD) techniques. We apply SIMD to a simple interior point method (IPM) that is capable of solving multiple similar linear programs in parallel. We compare our method against a non-SIMD linear program solver (CLP) and demonstrate significant speed-ups for some water resource simulations. Although our method is not competitive with the conventional solver for a single scenario, it is able to outperform the conventional solver when simulating multiple scenarios. Our results demonstrate that by exploiting modern CPU features, it is possible to achieve significant speed-ups for water resource simulations. We achieve a speed-up of 2.1 when using SIMD in comparison with a non-SIMD approach with 64 scenarios. More efficient simulation will allow practitioners to explore more scenarios, find more robust solutions, and use more complex models.