The US Bureau of Reclamation (USBR) has announced the winners of the Divide and Conquer Challenge, a competition aimed at enhancing the speed and efficiency of numerical models used in simulating hydraulics, sediment erosion, transport, and deposition in rivers and reservoirs.
As demands grow for analyzing large river systems and monitoring hydrological changes over time, computational simulation models have become increasingly vital. Recognizing the need for accelerated models that can deliver results to decision-makers more rapidly and cost-effectively, the Bureau of Reclamation organized the challenge.
Levi Brekke, Senior Advisor of Research and Development, expressed the agency's objectives, stating, "Reclamation is facing an increasing demand to look at large river systems and how the hydrology is changing over time, which involves using computationally intensive simulation models. The results of this competition show promise that we can accelerate these models and provide results to decision-makers in a quicker, more cost-effective manner."
The winners of the Divide and Conquer Challenge are as follows:
- First Place: Christophe Choquet, from Vanves, Ile de France, France, was awarded $115,000 for developing a GPU-accelerated Parallel Smoothed Aggregation algorithm, implementing the SRH-2D solver. By leveraging previous simulation time steps as input and performing calculations on the Graphics Processing Unit (GPU), Choquet achieved a remarkable speedup in the range of 30 to 60 times compared to the baseline.
- Second Place: Xiaofeng Liu, from State College, Pennsylvania, secured the second-place position and received $85,000. Liu's solution presented a unified code that operates in three modes: serial, MPI, and OpenMP. This versatility allows for a seamless transition from the original serial version to the parallel version of SRH-2D. Additionally, Liu utilized the high-performance linear equation system solvers in Hypre, designed for massively parallel computers.
- Third Place: Mahdi Esmaily, from Ithaca, NY, US, claimed third place and a prize of $45,000. Esmaily's solution implemented MPI technology and a custom-developed graph partitioning algorithm to parallelize all components of SRH. A dedicated library was also created, simplifying the parallelization process while ensuring backward compatibility.
- Fourth Place: Zhi Jian Wang, from Lawrence, KS, US, took fourth place and was awarded $10,000. Wang's team parallelized the tool on multiple CPU cores using domain decomposition and MPI. Their approach employed a preconditioned CGS linear solver with an incomplete LU decomposition.
These winning solutions will be incorporated into USBR's sedimentation and river hydraulics model to demonstrate their enhanced execution speed compared to the agency's baseline model.
To facilitate the competition, USBR partnered with the Federal Highways Administration Office of Innovation Implementation - Resource Center, NASA Tournament Lab, and Freelancer.com.
