Configuration Timings for the HYCOM Gulf of Mexico Scenario

Abstract

HYCOM, standing for HYbrid Coordinate Ocean Model, is a general circulation model for the oceans (see Large, et al., 1994; Bleck, 2002; Wallcraft, et al., 2009). The primitive fluid dynamics equations are solved for the horizontal components of the flow and a reduced-order K-Profile-Parameterization (KPP) model is solved in the vertical to treat convection, diffusion and turbulent mixing implemented as a calibrated implicit diffusion representation. The hybrid grid in HYCOM adapts to treat deep ocean and shallow coastal regions appropriately. The model includes surface driving terms, ice and salinity physics and treats all the differentiated ocean layers. As part of a project to study reduced precision computation and other computational approaches to accelerate computing HYCOM regional solution ensembles, this report describes the use of Diablo interactive graphics embedded in HYCOM to execute an extended campaign of HYCOM Gulf of Mexico simulations to time various parallel processing configurations that involve one or more HPC computing nodes and different configurations of Open MP parallelism within the Message Passing Interface parallel framework built into HYCOM. The metric for comparing the different parallel approaches is the core efficiency, that is bang for the buck, using the NRL 32-node system luigi where each node has 128 cores and 256 gigabytes of memory but where all 128 cores are not used in many of the configurations tested. The optimal core efficiencies are found when a single node is filled with a number of MPI tasks (ranks) that leaves enough cores available for effective OpenMP parallelism within each task.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 05, 2024
Accession Number
AD1223409

Entities

People

  • Clark D. Rowley
  • Gopal Patnail
  • Jay Boris
  • Keith S. Obenschain
  • Prasad G. Thoppil
  • Robert O. Rosenberg
  • Tommy G. Jensen
  • Yu Y. Khine

Organizations

  • United States Naval Research Laboratory

Tags

Readers

  • Computational Modeling and Simulation
  • Ocean-Atmosphere Mesoscale Modeling, Data Assimilation, and Flux Boundary Layers
  • Parallel and Distributed Computing.