Field Guide · concept

Also known as: HA

High availability (HA) is the practice of designing systems to keep running with minimal downtime, by removing single points of failure through redundancy, failover, and health monitoring.1

clients Load balancer health checks Primary active Standby promoted on failure replication no single component can take the whole service down
A load balancer health-checks a redundant pair and sends traffic to the active primary; when the primary fails its health check, traffic fails over to the replicated standby, which is promoted — so one dead component never means an outage.

Overview

Availability is often expressed in “nines”: 99.9% uptime allows about nine hours of downtime a year, while 99.999% (“five nines”) allows about five minutes. Reaching high numbers means that no single component — server, disk, network link, power feed — can take the whole system down.

The toolkit includes redundant hardware, load balancers that route around dead servers, RAID for disks, replicated databases, and automatic failover to a standby when the primary fails. Health monitoring ties it together: the system must detect a failure quickly, or redundancy sits idle while users see errors. The data center itself contributes with redundant power and cooling.

The nines

Each additional nine cuts allowable downtime roughly tenfold — and gets sharply more expensive to guarantee:

Availability Nickname Downtime / year
99% Two nines ~3.65 days
99.9% Three nines ~8.8 hours
99.99% Four nines ~53 minutes
99.999% Five nines ~5 minutes

The right target is an economic choice: chase only as many nines as the cost of downtime justifies, since redundancy, testing, and on-call all scale with the goal.

Where it fits

High availability is about staying up, while scalability is about handling more load; the two are related but distinct, and many techniques (replication, multiple servers) serve both. Orchestrators like Kubernetes automate failover for containerized apps. A hobby GopherTrunk node rarely justifies HA, but a monitoring site that must never miss a call would replicate capture nodes and back-end servers so any one can fail without an outage.

Sources

  1. High availability — Wikipedia, on availability, redundancy, failover, and the nines. ↩

See also