Skip to content
IT Atlas

Infrastructure · High Availability

High Availability

Designing systems to survive component failure without service interruption, by removing single points of failure and automating failover.

IntermediateUpdated 2026-09-01

Overview

High availability addresses component failure within a site or region; disaster recovery addresses loss of the site or region itself. Conflating them produces designs that survive a failed disk but not a failed data centre.

Availability is a property of the whole dependency chain. A clustered database behind a single firewall, a single switch or a single power feed is only as available as that weakest link.

How it works

  1. 01Redundancy is provided at each layer: power, cooling, network paths, storage controllers, hypervisor hosts and application instances.
  2. 02Clustering coordinates failover with quorum and witness mechanisms to avoid split brain.
  3. 03Load balancers and health probes distribute traffic away from failed instances; stateless application design makes this straightforward.
  4. 04Synchronous replication preserves zero data loss over short distances; asynchronous replication trades some data loss for distance.

Design considerations

  • Map every dependency and identify single points of failure explicitly, including licence servers, DNS and identity.
  • Decide the failure domain: rack, room, site, availability zone, region.
  • Test failover under load; graceful failover in a quiet maintenance window proves less than it appears.
  • Balance cost against measured business impact per hour of downtime.

Failure points

  • Quorum loss causing a cluster to shut down deliberately.
  • Both cluster nodes on the same hypervisor host or storage array.
  • Health probes that pass while the application is failing.
  • Certificate or licence expiry affecting all nodes simultaneously.

IT vs OT

OT availability is measured in production continuity and safety, and redundancy is often achieved with hot standby controllers and deterministic networks rather than clustered general purpose servers.

IT Atlas covers Information Technology. Operational Technology is covered separately at otatlas.ai.