Have more overview

[jhidding]

The course is sectioned:

Introduction - Benchmarking: What Why (more speed!)
Computing Pi - Threading/MP: How (low level)
Delayed - Data flow: How (human level: managing complexity)
Whatever and exercise.

• Computing Pi
• Threading/MP: needs the overview
• Delayed
• Dataflow

---

• Concepts
  • chunking
  • dependency diagram
  • natural parallelism
• Benchmarking
  • timing your code
  • checking memory consumption
• First make things fast
  • numpy vectorization
  • numba just-in-time
• Then make things run parallel
  • threading
    • OS dependent internal mechanism for running functions concurrently (i.e. fork)
    • - need to lift the GIL somehow
    • + low on resources
    • + shared memory
  • multiprocessing
    • Starts a new process (like from the shell) and sets up communication channels in other ways
    • - large overhead
    • - need for data serialisation
    • - complicates shared memory
    • - less flexible: for instance no lambdas
    • + circumvents the GIL