- Goal: answer some questions related to rolling out new asset pipeline for audio
    - What effect will changing timing of moving assets into AVL have on delivery times?
    - How many workers will we need for this to work?
    - Will we need to take another or additional approach?

- Approach: run some simulations based on past year of data and see what we learn.

1. Gather data:
    - Did some querying in snowflake to make an approximation of when assets would move, if new system had been in place
        - When was earliest encoding order for a particular release?
        - How many files and how many bytes would be moved?
        - Gathered for 12 months, 11/2019 through 10/2020
    - see
        - query.sql
        - data in data/ directory
        - sims/dac/input_data.py

2. Simulate:
    - Use simpy (https://simpy.readthedocs.io/) to simulate various combinations of parameters, notably worker count and bandwidth
    - Simulations recorded projected wait time (time for assets to be moved into local storage) for each release
    - Also did some process pooling and caching since can take a while
    - See
        - main.py
        - sims/dac/simulation.py

3. Visualize:
    - Use pandas (https://pandas.pydata.org/) for analysis, generating plots
    - See
        - sims/dac/plotting.py

More to do:

- Add noise for download speed
- Validate bandwidth assumptions
- Simulate outages
- Investigate spikes