Uptime describes how often a system remains operational. High availability is often used for a system that stays available more often than an equivalent single node. But a high uptime percentage alone does not tell users whether requests succeed quickly enough to be useful. section 6.10.2 section 6.10.1
Availability needs a request-level meaning. Under the CAP theorem, an available system returns a non-error response to every request, though the data may not be current. That definition has no response-time limit: a response arriving an hour later could qualify technically while being useless in practice. PACELC adds latency to the design trade-off, including when there is no network partition. section 6.2 section 6.10.1 section 6.3
Why the percentage needs context
Suppose an SLA promises 50% uptime and a database achieves 80%. It exceeds the promise, but calling it highly available would be misleading. Likewise, five nodes that are each available half the time might achieve roughly 80% write availability with a write quorum of two. That still leaves about one write request in five without a successful response. section 6.10.2
When evaluating an uptime claim, ask what counts as a successful request, how quickly it must complete, and what happens during a network partition. A system may preserve availability during a partition by returning older data; another may preserve consistency and reject some requests. Neither an uptime figure nor the phrase “high availability” settles those trade-offs. section 6.2 section 6.3 section 6.10.2
























