The point of availability is that the service provider tries to guarantee a certain level of it and clients can expect that or more. In some cases (depending on service contracts) a penalty fee or decreased subscription fee is the consequence of an unexpected downtime.
The quality of availability is measured in fraction of percents; for example, 99.99 percent or 99.999 percent which are spelled out as "four nines" and "five nines", respectively. These values are considered pretty good availability values, but there is a small trick in computing this value. If the provider has a planned downtime that is announced in advance; for example, the annual or bi-annual maintenance for water pipes in a town doesn't make the availability number worse. The availability is only measured outside the planned maintenance window.
Let's see three examples. All examples list the real uptime and downtime during a full year. In the first example, a theoretical service provider has no planned maintenance window. In the second example, the service provider has one-week planned downtime during the whole year. In the third example, there is one hour planned downtime per day.
Percent |
No planned downtime |
One week downtime per year |
One hour downtime per day | |||
---|---|---|---|---|---|---|
Uptime |
Downtime |
Uptime |
Downtime |
Uptime |
Downtime | |
80.000 percent |
292d |
73d |
285d 14h 24min |
79d 9h 36min |
279d 20h |
85d |
90.000 percent |
328d 12h |
36d 12h |
321d 7h 12min |
43d 16h 48min |
314d 19h 30min |
50d 4h 30min |
99.000 percent |
361d 8h 24min |
3d 15h 36min |
353d 10h 19min 12seconds |
11d 13h 40min 48seconds |
346d 5h 3min |
18d 18h 57min |
99.990 percent |
364d 23h 7min 26sec |
52min 34sec |
356d 23h 8min 36sec |
7d 51min 24sec |
349d 18h 9min 38sec |
15d 5h 50min 22sec |
99.999 percent |
364d 23h 54min 45sec |
5min 15sec |
356d 23h 54min 52sec |
7d 5min 8sec |
349d 18h 54min 58sec |
15d 5h 5min 2sec |
100.000 percent |
365d |
0sec |
357d |
7d |
349d 19h |
15d 5h |
The uptime and downtime listed for the first example in the preceding table can be interpreted easily. The provider is serving (or thinks it's serving) an uninterrupted service and the users expect that and rely on that. In real life, this kind of service can be the previously mentioned natural gas (for heating and cooking), tap water, and sewage systems. However, nothing has unlimited capacity. The sewage pipes have limited throughput and a big storm can bring so much rain that the pipes can get suddenly full and overflow. This is an unexpected downtime in service and is obviously a lot of trouble for everyone. Fixing it may take hours or if the pipes have broken in the meantime, days.
However, let's consider the 0.001 percent downtime for the "five nines" case. The users experience denied or delayed service only 5 minutes and 15 seconds in total (for example, 864 milliseconds every day) during the whole year, which may not be noticed at all. Because of this, the service is perceived to be uninterrupted.
The second and third examples in the table show that no matter what the provider does, there is a minimum downtime and the uptime is converging to the maximum that can be provided.
Let's see what the planned downtime means and what can be done to hide it. Let's take a theoretical factory and its workers. The workers operate on certain machinery and they expect it to work during their work hours. The factory can have different shifts, so the machinery may not be turned off at all, except for that one week of maintenance. The workers are told to have their vacation during this time window. If there is really no other downtime, everyone is happy. On the other hand, if there is downtime, it means lost income for the factory and wasted time and lower salary for the workers.
Let's look at the sum of the "one hour every day" downtime. This means more than two weeks in total, which is kind of surprising. It's actually quite a lot if added together. But in some cases, the service is really not needed for that single hour during the whole day. For example, a back-office database can have automatic maintenance scheduled for the night, when there are no users in the office. This way, there is no perceived downtime; the system is always up when the users need it.
Another example of this "one hour downtime every day" is a non-stop hypermarket. Cash registers usually have to be switched to daily report mode before the first payment on the next day; otherwise they refuse to accept further payments. These reports must be printed for the accounting and the tax office. Being a non-stop hypermarket, it doesn't actually close its doors but the customers cannot pay and leave until the cash registers are switched back to service mode.
18.188.200.46