Card 08/ 08
Recap: Back-of-the-Envelope Estimation
Requests, peak, storage, downtime and latency, each as one line of arithmetic.
- An estimate multiplies stated assumptions into the numbers a design depends on, right within a factor of a few (Estimating Load From Stated Assumptions).
- Requests a second = daily users × actions each ÷ 86,400, then × a peak factor of 2 to 5 (Estimating Requests a Second From Daily Users).
- Size for the peak, plus one spare, never for the average (Sizing Capacity for the Average Instead of the Peak).
- Storage = count × size × days kept, with the unit written beside every number (Estimation Example: Storage for a Photo App).
- Availability is a downtime budget: 99.9% allows 8.76 hours a year, and each extra nine divides it by ten (Availability in Nines).
- Latency climbs by orders of magnitude from memory to SSD to the data centre to another continent, so count the trips a request makes (Latency Numbers).
- System Design
- Performance
← → move through sections and cards