Memory optimization of a radiative heat transfer solver for fire simulations. Caratenuto, A., Zhang, P., Wu, B., & Zhao, X. In 2018 Spring Technical Meeting of the Eastern States Section of the Combustion Institute, ESSCI 2018, volume 2018-March, 2018.
abstract   bibtex   
© 2018 Eastern States Section of the Combustion Institute. All rights reserved. The effects of thermal radiation in large-scale fire systems are extremely important to system behavior. Multiphysics combustion solvers that account for radiative heat transfer are necessary to accurately predict large scale fire behavior. High-fidelity radiation solvers tend to be computationally expensive and require excessive memory resources. However, recent developments in the high performance computing field have radically changed computing architectures and have reduced memory-To-core ratios, such as in new many-core systems. In this study, A MPI-parallelized Monte Carlo radiative heat transfer solver for a pool fire simulation is optimized on an Intel© Xeon PhiTM (code name "Knights Landing", or KNL) CPU for more efficient memory access. This CPU features 64 physical cores and can process with up to 256 logical threads, and therefore possesses a high capacity for processing multi-Threaded applications. A line-by-line database of approximately 3 gigabytes is required to assess the optical properties. At any given instance, every entry in the database has the possibility of being accessed during the ray-Tracing process. Before the optimization, a copy of the database must be loaded for each MPI process when running with MPI in parallel. The memory capacity of the Knights Landing CPU limits parallelization to below 64 MPI processes per node. After the optimization, 64-core concurrency is enabled. The addition of more MPI processes increases the overall efficiency of the pool fire simulation by reducing the total time-To-solution. A pre-vaporized n-Heptane pool fire is simulated, and the results are compared before and after the optimization to verify the implementation. Going forward, larger-scale simulations that are not limited to one computer node will be performed and the performance will be evaluated.
@inProceedings{
 title = {Memory optimization of a radiative heat transfer solver for fire simulations},
 type = {inProceedings},
 year = {2018},
 identifiers = {[object Object]},
 keywords = {Fire,Many-core,Memory Optimization,Radiation},
 volume = {2018-March},
 id = {74335589-de21-3f86-9e14-31e90c9bf1da},
 created = {2018-07-14T14:09:36.062Z},
 file_attached = {false},
 profile_id = {39bbd100-cc15-33d2-8ece-ae4ef95a4169},
 last_modified = {2018-07-14T14:09:36.062Z},
 read = {false},
 starred = {false},
 authored = {true},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {© 2018 Eastern States Section of the Combustion Institute. All rights reserved. The effects of thermal radiation in large-scale fire systems are extremely important to system behavior. Multiphysics combustion solvers that account for radiative heat transfer are necessary to accurately predict large scale fire behavior. High-fidelity radiation solvers tend to be computationally expensive and require excessive memory resources. However, recent developments in the high performance computing field have radically changed computing architectures and have reduced memory-To-core ratios, such as in new many-core systems. In this study, A MPI-parallelized Monte Carlo radiative heat transfer solver for a pool fire simulation is optimized on an Intel© Xeon PhiTM (code name "Knights Landing", or KNL) CPU for more efficient memory access. This CPU features 64 physical cores and can process with up to 256 logical threads, and therefore possesses a high capacity for processing multi-Threaded applications. A line-by-line database of approximately 3 gigabytes is required to assess the optical properties. At any given instance, every entry in the database has the possibility of being accessed during the ray-Tracing process. Before the optimization, a copy of the database must be loaded for each MPI process when running with MPI in parallel. The memory capacity of the Knights Landing CPU limits parallelization to below 64 MPI processes per node. After the optimization, 64-core concurrency is enabled. The addition of more MPI processes increases the overall efficiency of the pool fire simulation by reducing the total time-To-solution. A pre-vaporized n-Heptane pool fire is simulated, and the results are compared before and after the optimization to verify the implementation. Going forward, larger-scale simulations that are not limited to one computer node will be performed and the performance will be evaluated.},
 bibtype = {inProceedings},
 author = {Caratenuto, A. and Zhang, P. and Wu, B. and Zhao, X.},
 booktitle = {2018 Spring Technical Meeting of the Eastern States Section of the Combustion Institute, ESSCI 2018}
}

Downloads: 0