Regions and Availability Zones

Regions

A region is a geographic area (for example eu-west-1 in Ireland, us-east-1 in Virginia) containing several data centres. Choose regions for:

• Latency: close to your users.
• Data residency and compliance: some data must stay in a country or the EU.
• Service availability and price: new services and instance types appear in some regions first, and prices differ between regions.

Availability zones

Each region has several availability zones (AZs): one or more data centres with independent power, cooling and networking, connected by fast low-latency links. Spreading an application across two or three AZs protects it from the failure of one data centre, which is the most common large failure.

def availability(single, copies):
    return 1 - (1 - single) ** copies          # works if at least one copy is up (independent failures)

single = 0.995                                 # one zone: about 44 hours of downtime a year
for zones in [1, 2, 3]:
    a = availability(single, zones)
    minutes = (1 - a) * 365 * 24 * 60
    print(f"{zones} zone(s): {a:.6%} available, about {minutes:7.1f} minutes down per year")

1 zone(s): 99.500000% available, about  2628.0 minutes down per year
2 zone(s): 99.997500% available, about    13.1 minutes down per year
3 zone(s): 99.999988% available, about     0.1 minutes down per year

Multi-AZ versus multi-region

Multi-AZ is the standard for production: cheap, low latency, and many managed services do it for you (RDS Multi-AZ, load balancers across zones). Multi-region protects against a whole-region failure but adds data replication, failover and cost; use it when the business needs it.