Day 48 results: costing a design before building it
Measured on: (machine, OS, Python version)
Output
Paste part 1 (capacity and the failure simulation), part 2 (the bill), part 3 (the lever) and the scoreboard here.
One sentence per prediction
- P1 (servers at 70% with an N-1 spare):
- P2 (monthly bill):
- P3 (dominant line’s share):
- P4 (CDN saving):
Capacity, in my own words
How Little’s Law turned a request rate into a server count, and why the honest number was bigger than the naive one:
Did the headroom earn its keep?
What the failure simulation showed: how the 70% fleet behaved as servers died versus the hot 95% fleet, and the moment the hot one ran away:
The dominant line
Which line dominated my bill, was it the one I predicted, and why do teams so often optimise the wrong one:
The lever
What one lever I pulled, how much it saved off the whole bill, and how that compared with the compute lever:
The number that matters
Servers needed: __ . Monthly bill: __ . Dominant line: __ at __%. Saving from one lever: ___%.