What is the availability rate?
The availability rate is simple to calculate: the time during which the service was available, divided by the total time of the observed period, multiplied by 100. If your site was unavailable for 30 minutes over a 30-day month (43,200 minutes), its availability is (43,200 − 30) / 43,200 ≈ 99.93%.
This figure depends heavily on how “available” is defined. Is a server that answers Ping but whose website returns a 500 error available? For the user, no. That is why the type of check used matters as much as the percentage itself.
The “nines” translated into minutes
People often talk about the “number of nines” to describe an availability level. Each additional nine divides the tolerated downtime by ten. Here are the orders of magnitude, calculated over a 365-day year and a 30-day month:
- 99%: about 3 days and 15 hours of downtime per year, i.e. 7 h 12 min per month.
- 99.5%: about 1 day and 19 hours per year, i.e. 3 h 36 min per month.
- 99.9%: about 8 h 46 min per year, i.e. 43 minutes per month.
- 99.95%: about 4 h 23 min per year, i.e. 22 minutes per month.
- 99.99%: about 53 minutes per year, i.e. 4 minutes 19 seconds per month.
Which availability target should you aim for?
There is no universal right number. A personal blog can live perfectly well with 99%. An online shop that makes its sales in the evening will be far more sensitive, especially if outages fall during peak hours. An API used by partners will often have precise contractual commitments.
Every additional nine costs more: redundancy, multi-zone hosting, on-call procedures. Before aiming for 99.99%, it is worth asking what an hour of downtime really costs your business.
Remember too that the average hides the distribution: ten outages of four minutes and a single outage of forty minutes give the same percentage, but not the same experience for your customers.
How to measure uptime properly
A reliable measurement rests on a few simple principles:
- Measure from the outside: a service that monitors itself cannot report its own network failure.
- Choose the right test: HTTP for a website or an API, TCP for a service port (database, mail), Ping for the reachability of a machine.
- Check often: with a check every hour, a twenty-minute outage can go completely unnoticed. High-frequency checks give a much more faithful picture.
- Keep the history: permanent archiving of incidents lets you calculate availability over any period and spot recurring problems.
- Track response time too: a site that takes ten seconds to respond is technically “available”, but your visitors won't wait for it.
Tracking your availability with LOXUp
LOXUp, published by LOXBOX, monitors your services over HTTP, TCP and ICMP Ping with high-frequency checks. Every incident is archived permanently, and a stability graph lets you observe a monitor's behaviour over 5 minutes, 1 hour, 24 hours or several days.
When something goes down, the authorised contacts on your team are notified by email, SMS or push notification. You therefore know not only your availability rate, but also the story behind every minute lost.
Key takeaways
99.9% availability means about 43 minutes of downtime per month. The right target depends on the real cost of an outage for your business. For the figure to mean something, measure from the outside, often, with the right type of test, and keep the history of every incident.
Going further: SSL monitoring · uptime monitoring · all features