1 of 1

In Situ Particle Tracing for MPAS-Ocean�With the ImPACTS Partnership (BER)

Scientific Achievement

We are developing GPU-accelerated Lagrangian particle tracing for MPAS-Ocean simulations, coupled in situ through ASCR workflow software.

Significance and Impact

Multiple co-designed optimizations and implementations enable efficient time stepping, cell locating, and velocity field interpolation for particle tracing. No code modification is needed to couple climate codes with particle tracing, making in situ data analysis a drop-in addition to existing codes.

In situ coupling of MPAS-Ocean with OSUFlow particle tracing through the LowFive library. OSUFlow and LowFive were developed under ASCR research funding and are maintained through the SciDAC RAPIDS Institute.

Technical Approach

GPU-accelerated particle tracing in OSUFlow is many times faster than the particle tracer included in MPAS-Ocean.
In situ coupling offloads analysis operations such as particle tracing outside of MPAS-Ocean, allowing climate scientists to focus on the simulation.
ASCR workflow software makes code integration easy and seamless.

PI(s)/Facility Lead(s): Tom Peterka, Orcun Yildiz (ANL), Dmitriy Morozov, Arnur Nigmetov (LBNL), Hanqi Guo, Han-Wei Shen (OSU)

Collaborating Institutions: Argonne and Lawrence Berkeley National Labs, The Ohio State University

ASCR Program: SciDAC

ASCR PM: Kalyan Perumalla

Publication(s) for this work: Peterka et al. LowFive: In Situ Data Transport for High-Performance Workflows. In IPDPS’23.

Code Developed: https://github.com/diatomic/LowFive

LOCAL LAB POC:

Tom Peterka

TALKING POINTS:

We are developing GPU-accelerated Lagrangian particle tracing for MPAS-Ocean simulations.
The particle tracing is coupled with MPAS-Ocean using ASCR workflow and data transfer software.
The particle tracer is many times faster than the one included in the MPAS-Ocean code.
Coupling the two codes together is seamless and requires no changes to the codes.

METADATA:

Name of the associated awarded project: SciDAC RAPIDS Institute

PI name(s): The RAPIDS lead PIs are Rob Ross and Lenny Oliker. The RAPIDS investigators involved in this work are Tom Peterka, Orcun Yildiz (ANL), Dmitriy Morozov, Arnur Nigmetov (LBNL), Hanqi Guo, Han-Wei Shen (OSU). The SciDAC ImPACTS partnership Lead PI is Luke Van Roekel.

Name of the program manager: SciDAC managed by Kalyan Perumalla. Supporting technology managed by ASCR research (Margaret Lentz and Hal Finkel)

CITATIONS:

Peterka, T., Morozov, D., Nigmetov, A., Yildiz, O., Nicolae, B., & Davis, P. E. (2023, May). LowFive: In Situ Data Transport for High-Performance Workflows. In 2023 IEEE International Parallel and Distributed Processing Symposium (IPDPS) (pp. 985-995). IEEE.