What does the gap between 99.982% and 99.995% mean in annual minutes of downtime? Reading data-centre tiers and SLA figures correctly.
When choosing a data centre, the first term you meet is “Tier”. The classification developed by the Uptime Institute shows how redundant a facility is and whether it can keep running even during maintenance. But a Tier is a design class, not a guarantee.
The classes
- Tier I: Single power and cooling path, no redundancy. Expected downtime about 28.8 hours a year (99.671%).
- Tier II: Redundant components but a single distribution path. About 22 hours (99.741%).
- Tier III: Concurrently maintainable; at least two distribution paths. About 1.6 hours a year (99.982%).
- Tier IV: Fault tolerant; all systems 2N+1. About 26 minutes a year (99.995%).
Reading the numbers
The gap between 99.9% and 99.99% looks small, but annual downtime falls from 8.8 hours to 52 minutes. That is why the SLA text matters: how monthly availability is calculated and which outages are excluded.
Uptime is not just the data centre
The real availability of your server depends on the weakest link in the chain of power and cooling, network, storage, virtualisation and your own application. A Tier III facility can still fall below 99.9% with a single network cable or storage node.
Which is right for you?
- Corporate websites, e-commerce, SaaS: a Tier III facility + multiple carriers + redundant servers is enough.
- Banking, payment infrastructure, critical public systems: Tier III + a geographically separate second site (active-active) or Tier IV.
Certificates and tier differences
Tier is awarded in three stages: design documents, constructed facility and operational sustainability. Asking your provider which stage they hold is the easiest way to filter out marketing language. ISO 27001 (information security) and ISO 22301 (business continuity) certificates show that processes are actually run.
All Ovabit locations in Türkiye are designed to Tier III and positioned to provide geographic redundancy between İstanbul and Ankara. See our data centres page for details.
What should you ask as a customer?
- The actual (not committed) uptime over the last 12 months and the list of unplanned outages
- Power and cooling redundancy: 2N or N+1?
- Carrier diversity and upstream capacity
- Maintenance window policy and notice period
- Whether an audit visit is possible
The answers to these questions give far more valuable information than the figures in a marketing brochure.