Los Alamos National Labs with logo 2021

Parallel IMC and Simple Load Balancing with the Data Server Model

Research highlighting a variant of the “data servers” model from neutronics that is used to allow particle work to be decomposed independent of spatial and physical data

Contact  

Parallel IMC and Simple Load Balancing with the Data Server Model

Implicit Monte Carlo can dominate compute time in multiphysics simulations. This computational demand is made worse when the workload is not balanced between parallel processes. The standard method decomposes the spatial domain and passes particles between parallel processes. In the standard method, the load imbalance is proportional to the particle work imbalance, which can be large when an energy-based source strategy is used. A variant of the "data servers" model from neutronics is used to allow particle work to be decomposed independent of spatial and physical data.

Summary of full paper (pdf)