aws_notes_questions.txt

1) Trade-off between on-demand price/time and spot price/time? At what point does on-demand win?
    * Add on-demand model columns to dataframe

2) Impose celiing on costs/times, how choose bid ratio (given dataset size)? *Discussion point: bid only affects runtime, cost is fixed - this assumes our resource demands dont affect market price, which we can't prove or disprove either way.

3) Plots of affordability vs computer ownership (using spot-model instead of static, by region...?)

- Can you rely on mean price? How realistic is the mean model (static)? Do you have to run simulations?
- What is an effective bidding strategy? Look at price variability (volatility, variance, IQR)
- How much does it cost to run a given pipeline (CPAC, FSL, custom?)
- How confident can you be in estimates from pricing model? Use spot model to get error bounds of static model
- Compare with on-demand price. When is it worth to run on-demand vs spot?

- Change word "done" to "performed"

- Scatter plot costs vs time
- Scatter variance vs cost
- Mean/var of history vs mean/var of cost/times, or med/iqr of each