Glossary
Storage high availability
Storage high availability is the ability of a storage system to keep data readable and writable through drive, server and network failures, and through planned maintenance and upgrades. It is designed around the second failure that arrives while the first is being repaired, and around faults that take out several components at once.
What high availability covers, and what it does not
In a large system, drives fail and servers are patched every week. High availability means applications keep working through those events without a recovery procedure. It is distinct from disaster recovery, which restores service after a site or its data is lost, and from backup, which keeps earlier versions. A deleted or encrypted object stays as available as any other. How likely data is to survive hardware loss over years is a separate measure, data durability.
Failure domains set the limits
A failure domain is a group of components that can fail together. Protection that tolerates two failures is only as strong as the separation between the pieces it protects: two pieces of the same object in one rack turn a single rack outage into two failures.
| Failure domain | Typical event | Placement that survives it |
|---|---|---|
| Drive | Media failure | Pieces on different drives |
| Server | Hardware fault, reboot for patching | Pieces on different servers |
| Rack or power feed | Switch or power distribution failure | Pieces in different racks |
| Site | Fire, flood, regional outage | Copies or pieces at other sites, as in multi-region storage |
Why the rebuild window matters most
After a drive fails, the system regenerates its contents from the remaining pieces, and until that finishes the affected data has less protection than designed. A second failure inside that window is the event that loses data. Drive capacity has grown much faster than drive speed, so a rebuild funneled into one replacement drive can now take a day or more. Designs that spread rebuild work across many drives, and that use erasure coding with enough extra pieces to absorb another failure, keep the window short and survivable.
What happens when sites lose contact?
If the link between sites fails, each side may conclude the other is down. A system that lets both sides keep accepting writes ends up with two diverging versions of the same data. Most designs prevent this with a quorum rule: only the side holding a majority keeps accepting writes, and the minority pauses. When a side is declared down for good, failover moves its work to the survivor.
How availability is stated
Availability is quoted as the percentage of time a service is usable over a period. At 99.99 percent, the allowance is about 53 minutes of downtime a year. The figures are easy to quote and hard to compare, because providers differ on what counts as an outage, whether slow responses count, and whether planned maintenance is excluded.
Frequently asked questions
Is high availability the same as fault tolerance?
Not quite. Fault tolerance usually means no interruption at all when a component fails, through fully duplicated hardware. High availability accepts a brief pause or reduced performance during a switchover in exchange for lower cost.