Document Type
Presentation
Publication Date
8-1-2014
Abstract
Hydraulic and water quality models have been developed and used over decades for water distribution system (WDS) analysis. The models prove to be powerful tools for engineers to gain systematic understanding on the WDS conditions and conduct technically-sound WDS management. However, it takes significantly amount of computation time to perform one simulation run for large system with tens or hundreds of thousands of pipes over an adequately long period of duration in order to gain good results for system hydraulic and water quality dynamics. In the meantime, WDS analysis solvers are not developed to take the advantages of available computing units, which are no longer homogeneous, but heterogeneous, including many cores of Central Processing Unit (CPU) and Graphics Processing Unit (GPU). Both CPUs and GPUs commonly co-exist in the personal computers, tablets and smart phones. WDS analysis models must be able to take the full computing powers offered by this heterogeneous computing paradigm. In this paper, we report a parallel computation architecture that combines task parallelization and data parallelization on both CPU and GPU for efficient WDS hydraulic and water quality analysis. The task parallelization on CPUs is implemented with multi-threading computing technique while the GPU parallelization is developed using OpenCL to ensure the portability of the parallelized solvers on various hardware vendors’ devices. With the parallelized WDS models, hydraulic and water quality simulations can be executed in parallel on CPUs, and in the same time the water quality analysis can be speeded up on GPU with massively parallel computing threads. This paper also presents the performance analysis of the parallelized solvers using the heterogeneous and portable parallel computing paradigm with CPU and GPU.
Comments
Session R16, Model Development and Computation Technologies