Redundancy

Redundancy means keeping spare parts, links and power supplies ready, so that when one of them fails, a spare takes over and the service keeps running. A part with no spare, whose failure stops everything, is called a single point of failure. Redundancy is the work of removing single points of failure.

Here is how it works in general. Engineers first list every part whose failure would stop the service: a power supply, a disk, a server, a cable, even a whole building. For each one, they add at least one spare. If a system needs N parts to run and has one extra, the design is called N+1. If it has a complete second set, it is called 2N. The system then watches every part. When one fails, the work moves to the spare automatically, often within seconds. This switch-over is called failover. The failed part is repaired or replaced while the spare carries the load. Finally, engineers test the failover regularly, because a spare that has never been tested may not work on the day it is needed.

Take, for example, a data centre's electricity. The building draws power from the grid, often through two separate feeds. Data centre power systems also keep batteries, called an uninterruptible power supply (UPS), which run the servers until the diesel generators start and take over the load. This usually takes less than a minute, and the batteries are sized to last several minutes in case a generator is slow.

Consider a fibre cable dug up during road works. Operators lay their optical transport links in rings, so the traffic between two cities is sent the other way round the ring until the cut is repaired.

For instance, enterprise storage systems keep extra copies of the data on different disks, so a failed disk loses nothing. A distributed SQL database goes further and keeps copies of the data in several cities, so that even the loss of a whole data centre does not stop the app.

Redundancy costs money. A 2N design buys two of everything, and the spare equipment sits idle on most days. So each business decides how much downtime it can accept. A payment system or a hospital needs far more redundancy than an office printer.

How is redundancy different from a backup? Redundancy keeps a service running when a part fails. But if a file is deleted by mistake, or locked by ransomware, the redundant copies change at the same moment, so they are damaged too. A backup is a separate copy taken at a point in time. It is used to bring back the data as it was before the damage, even if that takes hours.