Telephony SLA: how to calculate availability by year and by month
The SLA for a SIP trunk or an 0800 (toll-free) number is a simple calculation, but full of details that change the result: the length of the period, scheduled maintenance, what counts as an outage and the resolution of the measurement. This guide walks through the calculation and each of those details.
Summary
- Availability = (period − maintenance − downtime) ÷ (period − maintenance) × 100.
- 99.9% allows about 8 h 46 min of downtime in a 365-day year, and about 43 min in a 30-day month.
- In Brazil, telephony SLAs are usually measured over the year; the month serves as a reference and an early warning.
- An outage only enters the calculation if someone measured it: the more frequent the test, the more accurate the number.
- Losing visibility because of a problem in the monitoring itself is not a circuit outage and should stay out of the calculation.
| Period | 99% | 99.5% | 99.9% | 99.95% | 99.99% |
|---|---|---|---|---|---|
| 365-day year | 3 d 15 h 36 min | 1 d 19 h 48 min | 8 h 45 min 36 s | 4 h 22 min 48 s | 52 min 34 s |
| Leap year (366 days) | 3 d 15 h 50 min | 1 d 19 h 55 min | 8 h 47 min 2 s | 4 h 23 min 31 s | 52 min 42 s |
| 31-day month | 7 h 26 min 24 s | 3 h 43 min 12 s | 44 min 38 s | 22 min 19 s | 4 min 28 s |
| 30-day month | 7 h 12 min | 3 h 36 min | 43 min 12 s | 21 min 36 s | 4 min 19 s |
| Leap-year February (29 days) | 6 h 57 min 36 s | 3 h 28 min 48 s | 41 min 46 s | 20 min 53 s | 4 min 11 s |
| February (28 days) | 6 h 43 min 12 s | 3 h 21 min 36 s | 40 min 19 s | 20 min 10 s | 4 min 2 s |
| Week | 1 h 40 min 48 s | 50 min 24 s | 10 min 5 s | 5 min 2 s | 1 min |
| Day | 14 min 24 s | 7 min 12 s | 1 min 26 s | 43 s | 9 s |
How do I calculate an availability SLA?
Divide the time the circuit worked by the time it was supposed to work. As a formula: availability = (period − scheduled maintenance − downtime) ÷ (period − scheduled maintenance) × 100. The period is counted in minutes, using the real calendar: a year has 525,600 minutes (527,040 in leap years), and a month has between 40,320 and 44,640.
Running it backwards answers the most common question in a contract negotiation: how much downtime a target allows. Allowed downtime = (period − scheduled maintenance) × (100% − target). For 99.9% over a 365-day year, that is 525,600 × 0.001 = 525.6 minutes, or 8 hours, 45 minutes and 36 seconds.
The calculator and the table above use exactly this formula. The table assumes no scheduled maintenance; with maintenance, use the calculator.
How much downtime fits in each target?
The figures that show up most often in telephony contracts, over a 365-day year:
| Target | Per year | Per 30-day month |
|---|---|---|
| 99% | 3 d 15 h 36 min | 7 h 12 min |
| 99.5% | 1 d 19 h 48 min | 3 h 36 min |
| 99.9% | 8 h 45 min 36 s | 43 min 12 s |
| 99.95% | 4 h 22 min 48 s | 21 min 36 s |
| 99.99% | 52 min 34 s | 4 min 19 s |
Each extra “9” divides the allowed downtime by ten. The difference between 99.9% and 99.99% is the difference between a whole afternoon down and less than an hour in the year. Before signing up to a high target, ask how it will be measured: a target nobody measures is just a number in the contract.
What counts as downtime?
This is the part the contract needs to spell out, and often doesn’t. For a voice circuit, the situations usually treated as unavailability are:
- the call doesn’t connect: the carrier rejects it with a SIP failure code, such as SIP 503 or SIP 404, or doesn’t respond at all;
- the call connects with no audio, in one direction or both: to the caller, a silent 0800 is down;
- the wrong message answers: a swapped IVR, or the carrier’s recording saying the number does not exist;
- voice quality drops below the agreed level: some contracts treat this as partial unavailability, others as a separate target.
And what normally stays out of the calculation:
- scheduled maintenance, announced and within the agreed window;
- failures caused by the customer itself, such as a PBX switched off at a branch, when the contract is with the carrier and the fault is not on the carrier’s side;
- time when the monitoring lost visibility because of a problem of its own. A test that could not run says nothing about the circuit, and cannot be turned into unavailability.
Whoever measures has to separate these situations, and the report has to show them. In Monitor, a period in which the test could not run because of a problem on 2N5’s side comes out of the calculation and is shown separately.
Yearly or monthly SLA: which one applies?
Whatever the contract says. In Brazilian corporate telephony, the availability target is commonly yearly, with each month’s figures used as a reference. The two measures tell different stories:
| Yearly target | Monthly target | |
|---|---|---|
| What it tolerates | A bad month, if the year closes within the target | No month outside the target |
| Allowed downtime at 99.9% | 8 h 45 min 36 s in the year | 43 min 12 s in each 30-day month |
| When the penalty is calculated | At year end | Every month |
| What the monthly figure is for | Acting before the year blows through the target | It is the SLA itself |
An example: a trunk is down for 6 hours in March and not at all for the rest of the year. Against a monthly target of 99.9%, March missed (6 hours against the 44 minutes and 38 seconds allowed in a 31-day month). Against a yearly target of 99.9%, the year closed at 99.932%, within the target.
That is why Monitor’s reports show the last 12 months as the figure that decides the SLA, with the month and the last 30 days for reference.
How do I subtract scheduled maintenance?
Take the maintenance out of the period before the calculation, on both sides of the fraction. A 30-day month has 43,200 minutes; with a 4-hour maintenance window (240 minutes), the period that counts drops to 42,960. If the circuit was down for another 30 minutes outside the window, availability is (42,960 − 30) ÷ 42,960 = 99.930%.
Two things to watch:
- The window has to be recorded beforehand, with a start and an end. A “maintenance” declared after the outage is just an outage.
- Whatever runs past the window is an outage. If maintenance scheduled to end at 4:00 a.m. ended at 5:30 a.m., the extra hour and a half goes into downtime.
In Monitor, a maintenance window covers the chosen circuits at a scheduled time, and failures on those circuits during the window do not open an incident.
How do I measure real downtime?
With a test that sees what the customer sees, at a frequency that fits the target. Two common mistakes make the SLA figure worthless:
- Measuring the equipment, not the call. A PBX that is powered on and a trunk that is registered do not prove that a call connects with audio. Downtime is only real if the test is a real call.
- Measuring too rarely. If the test runs once an hour, a 40-minute outage can go unseen, and the SLA comes out better than it was. For a monthly 99.9% target, which tolerates 43 minutes, the interval between tests has to be much shorter than that.
At Monitor’s default pace, each circuit gets a test call every 15 minutes. When one fails, the call is repeated three times, every 2 minutes; four failures in a row open an incident, and two good calls in a row confirm recovery. That way the start and end of every outage are pinned down to within a few minutes, and an isolated blip does not enter the SLA.
What should I ask for in the carrier contract?
Before accepting a target, agree in writing on:
- the measurement period (year or month) and the time zone;
- what counts as unavailability: only calls that fail to connect, or also silent calls and quality below the threshold;
- how maintenance is announced and the maximum window;
- who measures, and how: the carrier’s measurement, yours, or an independent one;
- what evidence is accepted to open a dispute: the time, the number called, the SIP code and the Q.850 cause of each failure.
The last item carries the most weight when it is time to claim. An outage recorded with the SIP code and Q.850 cause of every test call is hard to dispute; an outage noticed because a customer complained is not.
What about customer service?
For SACs (the customer-service channels Brazilian law regulates) in regulated sectors, availability is not just a contract matter. Decree No. 11,034/2022 (Decreto nº 11.034/2022) requires access to the SAC to be available 24 hours a day, 7 days a week, through at least one channel, and makes phone support mandatory. For telecom providers, Anatel’s General Consumer Rights Regulation, Anatel Resolution No. 765/2023 (RGC), requires phone support to be available 24 hours a day. Measuring the availability of the 0800 is how you show those rules were met.
Frequently asked questions
What is the formula for an availability SLA?
Availability is the time the service worked divided by the time it was supposed to work, as a percentage: (period − maintenance − downtime) ÷ (period − maintenance) × 100. A trunk that was down for 60 minutes in a 30-day month, with no maintenance, had 99.861% availability.
How much downtime is 99.9% availability?
In a 365-day year, 8 hours, 45 minutes and 36 seconds. In a 30-day month, 43 minutes and 12 seconds. In a week, 10 minutes and 5 seconds. In a day, 1 minute and 26 seconds.
Yearly or monthly SLA: which is better?
It depends on the contract. A yearly target tolerates a bad month as long as the year closes within it; a monthly one penalizes every bad month. In Brazilian corporate telephony the target is commonly yearly, with each month tracked for reference and to act before the year blows through it.
Does scheduled maintenance count as downtime?
Usually not, if it was announced with the notice the contract requires and stayed within the agreed window. It comes out of the period before the calculation. Maintenance that overruns its window counts as an outage from the moment the window ended.
How do I measure the downtime of an 0800 number?
By calling it at short intervals and recording when calls stop connecting and when they start connecting again. Precision depends on the interval: with one call every 15 minutes and retries every 2 minutes after a failure, the start and end of an outage are pinned down to within a few minutes.