Dynamic Load Balancing on Heterogeneous Architectures: An airplane Aerodynamics Use Case
Presenter
DescriptionWe can affirm with quite a certainty that future Exascale systems will be based on heterogeneous architectures, including accelerators such as GPUs. We can also expect significant variability on the performance of the computing devices engaged in a simulation; due to the explosion of parallelism and to other technical aspects such as hardware-enforced mechanisms to preserve the thermal design limits. In this context, dynamic load balancing becomes a must for the parallel efficiency of any simulation code. In this paper, we present the dynamic load balancing strategy implemented in the computational mechanics code Alya. At the distributed memory level, this is based on an SFC mesh partition which is adjusted at runtime according to time measurements. Complementarily, other runtime mechanisms are used at the shared memory space. This strategy is applied for the efficient exploitation of heterogeneous CPU/GPU architectures. The assessment of the performance of all the proposed strategies has been carried out for airplane simulations on the POWER9 architecture accelerated with NVIDIA Volta V100 GPUs.
TimeTuesday, 6 July 202114:00 - 14:30 CEST
LocationErnesto Bertarelli
Session Chair
Event Type
Minisymposium
CS and Math
Physics
Engineering