99.99% uptime: what hot failover and replication really mean
Four nines is a number that gets printed on a lot of brochures. Here is what it takes to actually earn it — and the three words that separate organizations that recover in minutes from those that recover in days.
What 99.99% actually permits
99.99% availability allows about 52 minutes of unplanned downtime per year. Not per incident — per year. For comparison, a single failed server with a same-day hardware replacement typically costs 4 to 8 hours. One such event blows the budget for the next decade. So four nines is not something you achieve by being careful. It is something you achieve by assuming failure and designing around it.
Backup vs. replication
A backup is a point-in-time copy of your data, stored somewhere else. It answers the question "can we get our data back?" It does not answer "how long until we're working again?" — restoring a backup means provisioning new hardware or cloud capacity, reinstalling, reconfiguring, then copying the data back. That is measured in hours or days.
Replication is different. A replica is a second, live copy of the entire system — data, configuration, operating state — kept continuously in sync with the first. When the primary fails, the replica is already running. There is nothing to rebuild.
Hot failover
Hot failover is the mechanism that moves your users from the failed primary to the replica automatically. Done well, a clinic's front desk notices nothing: the records system pauses for a few seconds and continues. Done poorly — or not at all — someone has to notice the outage, diagnose it, make a decision, and manually switch over. By then the 52 minutes are gone.
Resilient systems design
Replication and failover protect against a server dying. They do not protect against a single internet connection, a single firewall, a single power feed, or a single person who knows the password. Resilient design means walking every layer of the stack and asking: if this one thing fails, what happens? Where the answer is "everything stops," you add a second path.
Then — and this is the step most organizations skip — you test it. You pull the plug on purpose, on a schedule, and you write down what happened. A failover that has never been exercised is a theory.
What to ask your current provider
If the answers are vague, that is itself an answer. Our free consultation includes exactly this review — a real analysis of your network and a written report on your strengths and weaknesses, yours to keep.