Availability is measured through uptime percentages and formalized in service level agreements that define guaranteed performance standards and the consequences of falling short. Key benchmarks like three nines (99.9%) and five nines (99.999%) translate abstract reliability goals into concrete, contractually enforceable targets.
Measuring Availability & SLAs
We want availability, but how do we measure this?
When we say something is highly available, it's a concept, a subjective idea that something is available most of the time. But we need to measure things, and so that's where downtime comes into play. Downtime is something we can measure. It's not a concept; it's something that we can actually record and measure.
Downtime comes in two different elements, as we mentioned before: a frequency, how often it happens, and the duration, how long that downtime is. So we can measure this and we can add this up to figure out what our total downtime is.
The reverse of that would be uptime. That's the time that our systems are up and running, and a lot of times we express that in a percentage form. The percent comes from the total time that the system is up divided by the total time it should be up. So that is a measurement of uptime.
Now that we have something that's measurable — downtime and uptime, uptime in a percentage form — we can actually start creating some sort of standards of what we are aspiring to achieve.
So an example of this is maybe we're trying to achieve a 90% uptime. That means that we only can have 36 days out of the year of being down, which is actually a lot. Most customers are not going to handle 36 days of downtime; they would get frustrated with you and leave. We call this a one nine, which is a pretty poor level of standard here.
We could increase it to 95%, which is still 18 days — that's a lot. 97%, which is 10 days. 98%, which is 7 and a half days. These are still a lot. 99%, we start 3 days of downtime, and this is not necessarily continuous. That would be pretty devastating, but if it was broken up a little bit, maybe that's okay. We call that two nines.
99.9% is a common standard that's out there. This says that you can be down eight hours out of the year, so this is a very common one, three nines. Then we could have four nines, 99.99%, that means you can be down 52 minutes out of the year. So that's pretty hard to achieve; that's getting at a level that is going to be much more difficult to do. And then five nines is 99.999%, that means you can be down five minutes out of the year, which I can tell you is a pretty hard target to get.
Then you can also see that you can continue to increase it to six, eight, and nine nines, which gets to some pretty crazy numbers there. You'll see a lot of systems strive for this three nines right here. That means eight hours out of the year you can be down, and that is a somewhat acceptable level. But the gold standard here is five nines. If you can get it down to only 5 minutes of downtime out of the year, that's kind of considered a premium spot. So five nines is one that you will see really common out there as a standard to strive for.
There are something called service level agreements, or SLAs. An SLA is somewhat of a guarantee, where you're going to promote that we are going to deliver our services at a certain level, maybe it's three nines, maybe it's five nines. If we don't deliver that certain standard at that level, there possibly is some sort of compensation that we will give you, because we haven't upheld our part of the bargain.
So even though five nines is something to strive for, three nines might be something that we put on our SLA, and that's our guarantee that we will at least maintain these three nines, and if not, we'll give you a portion of your bill back. So that is your service level agreements.
One thing that's added to a service level agreement is the duration of measurement. What I mean by that is that you need to know whether you're going to measure this in a year, in a quarter, in a month, or what is that duration.
So as an example, if you're striving for five nines, that might be 5 minutes in a year, which is a difficult target. But if you're striving for that in a quarter, that means you can only have 79 seconds of downtime in that quarter — that's five nines, the 99.999%, for that quarter, for that 3 months. If you're looking at a month, that means you can only have 26 seconds of downtime. It doesn't give you very much wiggle room at all.
Another thing that an SLA usually outlines is some sort of maintenance windows. During the maintenance window, you're going to be patching systems and updating different services, perhaps doing some upgrades to the code, and during that time there's more potential for downtime, and sometimes it can't be avoided. So you need to schedule some sort of maintenance window. The maintenance windows are times when you can tell your customers to expect maybe some outages. So during the maintenance window you would have a scheduled time that you could bring systems down for maintenance.
TechKnowSurge builds IT and cybersecurity professionals through hands-on, concept-first training built around real understanding — not memorization. Free interactive tools, structured programs, and 25+ years of real-world experience, all in one place.
Explore free tools and programs →