Learn
Lesson · Availability and Failure · The fraction

Nines and Downtime

Availability is a fraction of a period, and the downtime it permits is (1 − A) × period. Each further nine divides that downtime by ten while moving the percentage by less than a tenth of a point, which is why the two ways of quoting the same target sound so unlike each other. The period is part of the number: a downtime figure without one says nothing at all.

downtime = (1 − A) × period A is a fraction, never a percentage 99.9% → 43 min 48 s a month, 8 h 45 min 36 s a year 99.99% → 4 min 22.8 s a month, 52 min 33.6 s a year k nines ⟺ 1 − A ≤ 10⁻ᵏ a month here is 2 628 000 s
One subtraction, one multiplication, one stated period

Core ideas

What this lesson establishes

One equation read in both directions, and the convention that makes its answer reproducible from one page to the next.

1

Availability is a share of a period

A is the fraction of a period during which the service does its job, so 1 − A is the fraction during which it does not, and multiplying that by the length of the period gives the downtime the target permits. Nothing else is happening: downtime = (1 − A) × period is the whole of this lesson, and every other lesson on this course is what happens to A when several of them are put together.

2

A nine is a factor of ten in the downtime

Going from 99.9% to 99.99% moves the percentage by 0.09 points, which sounds like a rounding. It divides 1 − A by ten, and 1 − A is what you actually buy: 43 minutes and 48 seconds a month become 4 minutes and 22.8 seconds. The percentage is a bad scale for this quantity and the downtime is a good one, which is why the lab draws the ladder in minutes.

3

The period is part of the number

“Fifty-two minutes of downtime” is four nines over a year and barely three nines over a month; the same duration is a different promise on a different clock. This course fixes one table of periods, and a month in it is a twelfth of a 365-day year — 2 628 000 seconds — so every monthly figure on every page can be reconciled with every other.

The material

Nines, minutes, and the period they are measured over

What an availability is, what a nine is worth, and why the period has to be written down before the arithmetic starts.

Definition · Availability

The availability A of a service over a period is the fraction of that period during which it is serving. It is a number in [0, 1]; the percentage is a presentation of it and never enters a calculation. The complementary fraction 1 − A is the unavailability, and the downtime permitted by A over a period of length T is (1 − A) × T.

Both directions of that equation are used constantly. A target hands you A and you want the minutes; a downtime budget hands you the minutes and you want the A it corresponds to. They are the same equation rearranged, and the lab does both.

given A: downtime = (1 − A) × T given downtime: A = 1 − downtime/T A = 999/1000, T = 2 628 000 s downtime = (1/1000) × 2 628 000 = 2628 s = 43 min 48 s

Keep A as a fraction throughout. 999/1000 is exactly three nines and 0.999 is a double that is very slightly not; the differences this course cares about live in the fourth decimal place and further down, and a percentage rounded for display hides all of them. Every figure the lab prints is an exact fraction until the last digit.

Example · The same service, one nine apart

At 99.9% you may be down 43 min 48 s a month and 8 h 45 min 36 s a year. At 99.99% those become 4 min 22.8 s and 52 min 33.6 s. The percentages differ by 0.09000%; the minutes differ by a factor of ten, and it is the minutes that get paid for — in redundancy, in deployment discipline and in how fast someone has to be woken up.

Definition · Nines

An availability has k nines when 1 − A ≤ 10⁻ᵏ, taking the largest such k. This is a statement about the unavailability and not about how many 9 characters appear in the printed percentage.

The two readings come apart immediately. 99.95% shows two nines in its digits, but 1 − A = 0.0005, which is smaller than 10⁻³ and larger than 10⁻⁴, so it has three nines — and it permits 21 min 54 s a month, half of what 99.9% permits. Nines are a logarithmic scale, so “three and a half nines” is not a category the definition contains.

The period is not optional

A downtime is a duration and a duration only means something against a period. This course fixes one table and states it on every page: an hour is 3600 s, a day 86 400 s, a week 604 800 s, a month 2 628 000 s, a quarter 7 884 000 s and a year 31 536 000 s. The month is a twelfth of a 365-day year, which is the convention under which three nines is the 43.8 minutes quoted everywhere above.

Another convention is in circulation and it is also defensible: a 30-day month is 2 592 000 seconds, and three nines over that is 2592 s, or 43.2 minutes. Neither figure is wrong. What is wrong is quoting one of them beside an annual figure computed from the other, and that is why the period is written down first rather than assumed.

The other direction: a budget is an availability

Start from what you are willing to be down and the same equation hands back the target. This is usually the more honest order, because a team can say how long an outage may last long before it can say which nine it wants, and the arithmetic converts one into the other without any negotiation.

Example · A budget that does not quite buy a nine

Someone rounds 43.8 minutes up and writes “44 minutes a month” into a document. That budget is 1 − 2640/2 628 000 = 10939/10950, or 99.89954%, and by the definition above it has two nines rather than three. The rounding went the wrong way across a boundary. Set the lab to work from a budget and move the slider between 43 and 44 minutes to watch the count change.

Interactive

Nines are minutes

Downtime is (1 − A) × period, and each nine is a factor of ten

Nines and downtimeeach further nine is ten times less of it, not a hundredth of a per cent more
the nine you chosethe ladderyour budget

Method

Turning a target into minutes

Four lines, and the third one is the only arithmetic. The other three are what stop the arithmetic being done on the wrong number.

1

Write the availability as an exact fraction

99.9% is 999/1000. Do not carry it as 0.999, and do not carry the percentage into the multiplication: the factor you need is 1 − A, and here that is exactly 1/1000.

2

Name the period and put it in seconds

Write down which period you are quoting and its length before anything else. A month on this course is 2 628 000 s. If a different convention is in use, say so in the same sentence as the answer.

3

Multiply, then read the product as a duration

(1/1000) × 2 628 000 = 2628 s. Convert once, at the end: 43 minutes and 48 seconds. Converting earlier introduces a rounding into a quantity whose interesting digits are at the far right.

4

Check the answer against the ladder

One more nine must give a tenth of the downtime and one fewer must give ten times as much. If your answer does not sit on that ladder, the error is almost always a period mismatch — a monthly figure compared with an annual one.

Nines, minutes and periods

Score 0 / 0

99.95% into minutes, and 21 minutes 54 seconds back into a fraction

A target that is not a round nine, so the fraction has to be carried rather than recognised.

A = 99.95% = 9995/10000 = 1999/2000 1 − A = 1/2000 a month, 2 628 000 s downtime = (1/2000) × 2 628 000 = 1314 s = 21 min 54 s a year, 31 536 000 s downtime = (1/2000) × 31 536 000 = 15 768 s = 4 h 22 min 48 s and back again A = 1 − 1314/2 628 000 = 1 − 1/2000 = 1999/2000 = 99.95% how many nines 1 − A = 0.0005 ≤ 10⁻³ so k is at least 3 0.0005 > 10⁻⁴ so k is not 4 99.95% is three nines, and it permits half the downtime of 99.9%

The check at the end is the part worth keeping. Two digits of the percentage moved and the count of nines did not; the downtime halved. Anyone reading “99.95” as “two nines and a bit” has compared the printed characters instead of the unavailability.

For a faded rehearsal, take 99.5%. The supplied first move is 1 − A = 1/200. Produce the downtime per week and per quarter, say how many nines it has by the definition rather than by looking, and then state the monthly budget in minutes that would buy exactly one more nine. Check all three against the lab before opening the quiz, and say which period each of your figures belongs to.

Common mistakes

1
Comparing two targets by subtracting their percentages99.99 − 99.9 = 0.09 is arithmetic about the wrong quantity. The thing that changed is 1 − A, and it changed by a factor of ten. Any argument of the form “that is a tiny improvement” based on the percentage gap has measured the availability where it should have measured the outage.
2
Quoting a downtime with no period attached“We were down 52 minutes” is four nines against a year and worse than three against a month. A duration is half a number; the period is the other half. Write both, every time, and prefer to write the period in seconds so that nobody has to guess whether your month has 30 days or 30.4167 of them.
3
Counting the nines that are printed rather than the ones that are bought99.95% displays two nines and has three. 99.5% displays two and has two. The definition looks at 1 − A and asks which power of ten it fits under, which is a question about the size of the outage rather than about the spelling of the target.

Check your understanding

✓
Finish when a target availability and a downtime budget feel like one number written two ways.

You should be able to convert either into the other over any of the six periods without looking anything up, say how many nines an availability has by the definition rather than by eye, and state the period alongside every duration you quote.

Note: This lesson takes the availability as given. Where it comes from is the next question, and it has an answer with two levers in it: a machine is available in proportion to how long it runs between failures and how fast it is repaired. “MTBF and MTTR” shows that those two levers are worth exactly the same amount, and that one of them is usually far cheaper to pull.

Kept in this browser only, for you. Nothing is sent anywhere and nothing checks it. Carry it to another device.

Feedback and recommendations

Something to change about this lesson? Write it down, tick it off when it is handled, and export the list when you want to act on it. Kept in this browser only: nothing is sent anywhere, and nobody is notified.

No recommendations yet for this lesson.