Hands-on design challenge

Challenge 16: Losing a zone

The situation

The web app from the primer’s design runs in one AWS Region, and must keep serving when a whole availability zone fails. Users read far more than they write. Most reads are for images, files and pages that are the same for every user, which a CDN can serve, and the rest mostly for a small set of popular content, which the memory cache holds.

Traffic

20,000 requests a second.

OperationShareSteps
Read95%Read data
Write5%Write data
  • Read95%Read data
  • Write5%Write data

Data

A CDN can answer 80% of reads at an edge. A cache answers 90% of the reads that reach it.

What you are proving

A design survives losing a zone when the zones that remain already hold enough capacity to carry the whole load, with nothing to launch.

Rules

  • A zone fails as a whole, taking every instance in it.The data centers of a zone share its power and networking, so the lab takes from each tier the instances in its largest zone.
  • No instance is launched while the zone is down.Replacing lost capacity depends on launches working during the incident, and requests are dropped until they do; the lab counts the loss before any replacement.

Builds on

Targets

  • Serve all 20,000 requests a second
  • At least 99.99% available, with no single point of failure
  • At most 5,300 USD a month
  • Any tier loses one instance: nothing dropped, no tier over 80% busy
  • One availability zone is lost: nothing dropped, no tier over 80% busy

Hints