Scalability Study of Particle Method with Dynamic Load Balancing
We introduce an efficient load-balancing algorithm for particle method (Particle Blast method and Corpuscular Particle method) in LS-DYNA®. Load-balancing is achieved by dynamically adaptively using RCB to evenly distribute workload to processors. Numerical tests demonstrated that with reformulated parallel scheme, for PBM, the speedup for an airblast problem can be 20~30 times or more when using 128~192 cores; for CPM, several times speed up can be achieved for a curtain airbag simulation.
https://www.dynalook.com/conferences/15th-international-ls-dyna-conference/blast/scalability-study-of-particle-method-with-dynamic-load-balancing/view
https://www.dynalook.com/@@site-logo/DYNAlook-Logo480x80.png
Scalability Study of Particle Method with Dynamic Load Balancing
We introduce an efficient load-balancing algorithm for particle method (Particle Blast method and Corpuscular Particle method) in LS-DYNA®. Load-balancing is achieved by dynamically adaptively using RCB to evenly distribute workload to processors. Numerical tests demonstrated that with reformulated parallel scheme, for PBM, the speedup for an airblast problem can be 20~30 times or more when using 128~192 cores; for CPM, several times speed up can be achieved for a curtain airbag simulation.
15-A_Blast_019.pdf
— 783.7 KB