25. Approaches to parallel computing in CFD

This lecture starts with a brief introduction to, and a history of parallel computing. Next, the most widely used parallelization concepts suitable for application in CFD are presented, followed by a description of factors affecting the efficiency of parallel computing.
Domain decomposition in space is the most widely used approach to parallel computing in CFD and is therefore described in more detail. Domain decomposition in time is not widely used, but it is described because it can be combined with domain decomposition in space to increase the efficiency of long-lasting transient flow simulations.
Analysis of the efficiency of parallel computing is also addressed, considering the necessary modifications of the solution algorithm, the additional time for communication between processors, and the load balancing effects.
In recent years, CFD methods have been also adapted to run on graphic cards. Vendors of commercial CFD software are also following this trend by porting the codes to computing systems based on graphic cards. At the end, the advantages and disadvantages of the various approaches are highlighted in some practical application examples.
You can download this video, the transcript of its soundtrack, and the PowerPoint presentation file from this site: https://sie.ag/6a3r6J
You are also encouraged to participate in the discussion by posting comments, questions, and suggestions for improvements in the dedicated space for "The Perić Lectures on CFD" at the above site.