A Path from Serial Execution to Hybrid Parallelization
for Learning HPC
Author/Presenters
Event Type
Workshop
Education
TimeMonday, November 13th9:25am -
9:45am
Location505
DescriptionParallel and distributed computing are becoming
necessary in almost all aspects of computation. Due to
this growing demand, curriculum initiatives have been
developed for integrating parallel and distributed
computing into traditional undergraduate computer
science programs. However, adoption has been slow
resulting in many students lacking proper training for
parallel and distributed computing. Two potential
barriers for slow adoption are a deficiency in example
programs that step students through the processes of
parallelizing serial code, and the inaccessibility of
dedicated machines to run highly parallel programs at
scale within the confines of a course schedule. We have
developed course material using a simple two-dimensional
Lattice-Boltzmann Method Computational Fluid Dynamic
simulation to walk students though shared memory
parallelism, distributed memory parallelism, and hybrid
parallel execution. We also created a custom
mini-cluster comprised of 16 credit-card sized compute
nodes, with a total of 288 cores, as an inexpensive
solution for testing the scalability of different
parallel models that can be deployed in a classroom
setting.
