https://haojian.github.io/DSC102SP24/static_files/presentations/7ParallismBasics.pdf
https://haojian.github.io/DSC102SP24/static_files/presentations/8ParallismDataAccess.pdf
Quantifying Benefit of Parallelism
Idle Times in Task Parallelism
Calculating Task Parallelism Speedup
Task Parallelism in Dask
- Ops on data struct. are NOT executed immediately
- Triggered manually, e.g., compute()
- Dataflow graph / task graph is built under the hood
Dask Workflow
- Optimize
- Serialize task graph
- Communicate to scheduler
- Scheduler divides up work
- Workers begin their partitions
Possible Bottlenecks

Best Practices for Task-Par. Dask