A single data center is a single point of failure. However well it is built, if that facility loses power or connectivity, everything inside it goes down together. The fix is a second site. The question is how far away it should be.
The clearest public answer comes from Amazon Web Services. AWS describes its Availability Zones as meaningfully distant from each other, up to 60 miles (about 100 km), to prevent correlated failures, but close enough to use synchronous replication with single-digit millisecond latency. The zones are designed to avoid simultaneous impact from shared fate scenarios such as utility power, water disruption, fiber isolation, earthquakes, fires, tornadoes, or floods. They are designed to be supplied by different power substations and are linked by high-bandwidth, low-latency networking over fully redundant, dedicated metro fiber.
That is the design problem in one paragraph. Far enough that one event cannot take both sites. Close enough that the two sites behave as one.
There are three tiers, and they are different products. A single site has no redundancy. A metro pair, which we build with the second site 20 to 50 miles from the first, uses synchronous replication and survives the loss of a whole site. A cross-region copy, typically hundreds of miles away, survives a regional disaster, but replication at that distance is usually asynchronous, which can mean some data loss and a longer recovery. Metro high availability and disaster recovery are not the same thing. Calling a metro pair “disaster recovery” overstates it. Calling a cross-region copy “high availability” understates the outage. The distance decides which product you are buying.
Vendor limits show how far synchronous replication can stretch. Nutanix Metro Availability stretches a compute and storage cluster across two sites, with near-zero recovery time and a recovery point of zero, using synchronous replication. Nutanix describes zero RPO for sites within 400 km or under 5 ms of latency. NetApp MetroCluster mirrors data across two sites in the same geographical region, up to 200 km over Fibre Channel and 300 km with Fibre Channel over IP, and can switch over to the other site, manually or automatically, if a whole site goes down. NIST’s contingency planning guide (SP 800-34 Rev. 1) gives no fixed distance. It says an alternate site should be in a geographic area unlikely to be affected by the same hazard as the primary site.
Why we use 20 to 50 miles comes down to shared fate. That range is our design choice, not an industry standard. Sites that are very close can share a substation, a flood plain, or a fiber route, so a single event can take both. As distance grows, latency grows with it, and synchronous replication eventually stops being practical. Between 20 and 50 miles, the two sites can be put on different substations and different fiber routes, and they are close enough to stay in sync. That is the range we use. It sits inside the envelope AWS publishes for its own Availability Zones, which are up to 60 miles (about 100 km) apart.
That is the product we sell. Edge AI mirrors the workload at a second site 20 to 50 miles from the first, at 25 percent off the second site for the first 36 months, then full price. We call it Metro High Availability, because that is what it is. The mirror is live and operational, running in a serial path to end users, not just backup or standby. That is the difference between a second site that works and a second site that is a promise. We offer it wherever we build.
If you are buying high availability, ask three questions. How far apart are the two sites. Are they on different substations. And is the second site live, or just standby. If a provider cannot answer all three, they are selling you a word, not a second site.
High availability is a distance and a live second site. It is not a word on a brochure.
Sources
- Amazon Web Services, Availability Zones, AWS Fault Isolation Boundaries whitepaper: Availability Zones in a Region are meaningfully distant, up to 60 miles (about 100 km), to prevent correlated failures, but close enough for synchronous replication with single-digit millisecond latency; designed to avoid simultaneous impact from utility power, water disruption, fiber isolation, earthquakes, fires, tornadoes, or floods; supplied by different power substations; interconnected over fully redundant, dedicated metro fiber. https://docs.aws.amazon.com/whitepapers/latest/aws-fault-isolation-boundaries/availability-zones.html
- Nutanix, Metro Availability: stretch clustering across two sites, near-zero recovery time and a recovery point of zero, synchronous replication. https://next.nutanix.com/how-it-works-22/metro-availability-37294
- Nutanix, DR multi-site recovery (blog, May 4, 2020): zero RPO for sites within 400 km or less than 5 ms latency. https://www.nutanix.com/en_gb/blog/nutanix-dr-multi-site-recovery
- NetApp MetroCluster tutorial (FlackBox): redundancy across two buildings in the same geographical region, synchronous mirroring, failover if a whole site goes down, up to 200 km over Fibre Channel and 300 km with Fibre Channel over IP. https://www.flackbox.com/netapp-metrocluster-tutorial
- National Institute of Standards and Technology, Contingency Planning Guide for Federal Information Systems (SP 800-34 Rev. 1, May 2010): an alternate site should be in a geographic area unlikely to be affected by the same hazard; no fixed distance is given. https://nvlpubs.nist.gov/nistpubs/legacy/sp/nistspecialpublication800-34r1.pdf
