1. Practical Course on Parallel Numerical Methods
The first practical course on parallel numerical methods took place online from 27.08.2020 - 07.08.2020 due to the corona pandemic.
Course instructor: Markus Scherg (Chair of Parallel and Distributed Systems) Additional supervisor: Thomas Rau (Chair of Scientific Computing)
Participants: Four students of the master's program "Scientific Computing"
The importance of parallel methods
The calculation and simulation of real problems becomes increasingly complex due to their more complex structure. Therefore, parallel implementation or the development of algorithms that can be implemented in parallel is gaining more and more importance nowadays. The distribution of the workload to several cores or processors is in the foreground, in order to significantly reduce the total time required.
The course started with some theoretical basics about parallel programming and the classification of existing finite element software regarding their ability to perform parallel computations. With regard to parallelizable algorithms such as Schwarz methods, these software packages have been examined for domain decomposition. The goal of the course was to implement a parallel PCG (preconditioned conjugates gradient) method. After a short introduction to common libraries like OpenMP, MPI and the architecture and functionality of cluster computers, the participants of the Bayreuth Cluster were allowed to test and evaluate their programs.
In this practical course, students should learn the parallel implementation of already known algorithms that have been analyzed in previous courses. All students managed the implemention of the parallel PCG method. The students could present their results and findings in the final lectures on 07.08.2020. As it unfortunately turned out, the course program was very extensive, so that the extended task could only be worked on very sporadically. This was the implementation of a parallel domain decomposition method according to Schwarz.