High Availability Is Not a Disaster Recovery Plan

Share
High Availability Is Not a Disaster Recovery Plan
Photo by Domaintechnik / Unsplash

High availability is one of those terms that gets used rather loosely in the hosting industry. Put three servers in a cluster, add redundant storage and a couple of network switches, and suddenly the entire platform is described as highly available. Technically, that might be correct. Operationally, there's a rather important distinction that often gets overlooked: high availability keeps services running when individual components fail, while disaster recovery gets them running again when the environment itself is no longer available. They're related disciplines, but they're certainly not interchangeable.

Consider a virtualisation cluster with three physical hosts. One host suffers a hardware failure, and the affected virtual machines restart on the remaining nodes. Assuming sufficient spare capacity, shared storage availability and correctly configured HA policies, the interruption might be relatively brief. That's high availability doing its job. Now imagine the storage system suffers a major fault, a networking change isolates the entire cluster or the datacentre experiences an extended power failure. The fact that there are three physical servers doesn't solve any of those problems.

Understanding the failure domain

Every infrastructure design has failure domains: the components or services that can fail together because they share a dependency. Those dependencies aren't always obvious. Two servers might have independent power supplies but connect through the same switching infrastructure. Storage might be replicated between multiple nodes, while the entire cluster still depends on a single management or network configuration. Even geographically separate environments can share external DNS providers, identity systems or network carriers. A genuinely resilient design starts by identifying these relationships, rather than simply counting redundant components on a diagram.

Recovery is a business requirement

The conversation becomes much more useful when businesses establish two figures. Recovery Time Objective (RTO) is how long a service can reasonably remain unavailable. Recovery Point Objective (RPO) is how much data loss the business can tolerate, expressed as time. An accounting application that can survive several hours of interruption has different requirements from a transactional platform processing thousands of operations every minute. Both might benefit from high availability, but their disaster recovery requirements could be completely different.

There's a cost implication, too. Achieving a recovery time of minutes across geographically separate locations is a very different engineering exercise from restoring an application within four hours. Neither is automatically the better solution. What matters is whether the design matches the business requirement and whether the people operating it understand what the stated targets actually mean.

Design for what can actually go wrong

At KORE, we approach resilience as an infrastructure-wide consideration. Compute, storage, networking, management access, backup and recovery all need to be considered together. A virtual machine restarting automatically is useful, but knowing where the application will run if the entire site becomes unavailable is considerably more important. That also means testing. A disaster recovery plan that has never been exercised is a set of assumptions, not a demonstrated capability.

The objective isn't to build the most complicated infrastructure possible. It's to understand what can fail, what happens when it does and how the business returns to normal. High availability is part of that answer. It isn't the whole answer.

Read more

Beyond the LLM: Engineering Secure, Reliable AI Agents for Enterprise Infrastructure

Beyond the LLM: Engineering Secure, Reliable AI Agents for Enterprise Infrastructure

Building a capable artificial intelligence model is one challenge. Integrating that intelligence into business-critical infrastructure, while maintaining security, reliability and operational control, is an entirely different engineering problem. The rapid advancement of large language models has created considerable enthusiasm around the possibility of intelligent automation. Modern models can interpret

By KORE Pulse