Session 3: Fundamentals of parallel processing
Session Rules
- Chat questions/comments take first priority - Chat your question/comments either to everyone (preferred) or to the chat moderator (Ryan Lucas) privately to have your question/comment read out loud anonymously. We will answer chat questions first and call on people who have written in the chat before we take questions from raised hands.
- Share your video when speaking - If your internet plan/connectivity allows, please share your video when speaking.
- Keep yourself on mute - Please mute yourself when not speaking.
Learning objectives
The overall objective of this session is to provide a foundation in parallel processing terminology and HPC-usage basics before applying these concepts in the upcoming tutorials. Please see the links listed in Session 11 for additional parallel processing training opportunities.
- Define an ‘embarassingly parallel’ problem
- Apply a parallelization approach flowchart to a geospatial problem
- Describe the basic SLURM job submission parameters
- Identify the appropriate Ceres or Atlas partition for a job
Agenda
This session will be a lecture with slides covering the following content:
- Trade-offs in parallel processing: effort, time, communication
- How to approach parallelization: a geospatial example
- Determine where in your workflow the most time is spent
- Flow chart: parallelization approaches
- Choosing number of cores
- SLURM jobs, nodes, and partitions
Watch a recording of this presentation.