Prev: latency-and-throughput Next: caching

Availability

• Customers wouldn’t be happy if a website was down — we would lose money.

• Some products are life or death, like airplane systems.

• Cloud providers like AWS and GCP also have high availability (99.95%)

• We measure availability by how much uptime they have.

• 99% - Two Nines - 3.65 days of downtime

• 99.9% - Three Nines - 8 hours of downtime

• 99.99% - Four Nines - 1 hour of downtime

• 99.999% - Five Nines - 5 minutes of downtime (High Availability)

• Service Level Agreement (SLA) are a guarantee of how much uptime is required.

• Service Level Objective (SLO) are similar.

• How do we increase availability?

• eliminate single points of failure.

• Make sure there’s redundancy.

  • If you have one server, have multiple.
  • If you have multiple servers, get more load balancers.

• Active redundancy is like leader election, when a node crash reconfigures itself.

Prev: latency-and-throughput Next: caching