“99.9% uptime” appears on almost every connectivity quote, and it is read by almost everyone as “it will not go down”. It is a precise arithmetic claim with a specific allowance for downtime built into it — and the clauses around it decide whether the number means anything at all.
What the nines are worth, in minutes
Availability is downtime as a fraction of a measurement period. Taking a 30-day month and a 365-day year:
| Availability | Allowed downtime / month | Allowed downtime / year |
|---|---|---|
| 99% | 7 h 12 min | 3.65 days |
| 99.5% | 3 h 36 min | 1.83 days |
| 99.9% (“three nines”) | 43 min | 8 h 46 min |
| 99.95% | 21 min 36 s | 4 h 23 min |
| 99.99% (“four nines”) | 4 min 19 s | 52 min 36 s |
| 99.999% (“five nines”) | 26 s | 5 min 15 s |
Two things fall out of that table immediately. First, the gap between 99% and 99.9% is enormous — a factor of ten, and the difference between “a working day of outage a year” and “an evening”. Second, 99.9% still permits a single 43-minute outage every month without any breach at all. If a 40-minute outage during business hours would be a serious event for you, three nines is not the promise you need.
What counts as downtime
An SLA has to define the failure it is promising against, and the definitions vary more than the percentages do.
- Total loss only, or degradation too? Many agreements count downtime only when the circuit is completely unavailable. A link that is up but delivering 10% of its capacity with heavy packet loss may not be “down” under the contract, while being unusable in practice. The stronger agreements define a performance floor — throughput, latency, packet loss — below which the service counts as unavailable.
- When does the clock start? At the moment of failure, or when you log a ticket? If it is the latter, an outage that begins at 2 AM costs you the hours until someone notices.
- Who measures? The provider’s own monitoring is the usual answer. Independent monitoring on your side is worth having if the service is important — you cannot dispute a figure you have no record of.
The exclusions, which are the real document
Every SLA carves out events outside the provider’s control. Most of these are reasonable. All of them reduce the promise.
- Scheduled maintenance. Standard, and usually excluded entirely. Ask for the notice period, the permitted window, and an annual cap on maintenance hours — an unbounded exclusion is not an exclusion, it is a loophole.
- Customer premises equipment and power. If your building loses power and the ONT dies with it, that is not the provider’s outage. Fair — and a reminder that your own UPS is part of your availability, not theirs.
- Third-party last mile. Where the final leg is delivered over another operator’s infrastructure, check whether the SLA still applies end to end or stops at the handover.
- Force majeure and civil works. Flooding, cyclone, road-widening that severs a duct. Real in any Indian city with an active monsoon and continuous roadworks.
- Anything upstream of the provider. No ISP can promise that a service you use elsewhere on the internet stays reachable.
MTTR usually matters more than the percentage
Availability tells you how much downtime is permitted in aggregate. Mean time to restore tells you how long any single outage lasts — and for most businesses, that is the number that actually hurts. Twelve four-minute outages a year and one fifty-minute outage both fit inside 99.99%; they are not remotely the same experience.
A serious agreement states a target MTTR, an escalation path with names or roles and timeframes, and whether the clock runs around the clock or only during business hours. If a quote has an impressive availability figure and no restoration commitment, the availability figure is decorative.
The remedy, and its limits
SLA breaches are almost always remedied with a service credit — a percentage of the monthly fee, refunded or applied to the next invoice, usually capped at 100% of one month’s charge. Read that honestly: the compensation for a bad outage is a discount on the line, not compensation for what the outage cost your business. An SLA is a statement of intent backed by a modest financial incentive. It is not insurance.
Also check how a credit is claimed. Many agreements require the customer to submit a claim within a short window — often 7 to 30 days — and pay nothing automatically. A credit nobody claims is a credit the provider never pays.
Separately from any commercial SLA, TRAI’s Standards of Quality of Service of Access (Wireline and Wireless) and Broadband (Wireline and Wireless) Service Regulations, 2024, in force since 1 October 2024, require compensation to subscribers for prolonged network outages — rent rebates for postpaid customers and validity extensions for prepaid ones — and set graded financial penalties on providers for breaching quality benchmarks. That is a regulatory floor beneath whatever your contract says, not a substitute for reading it.
Your real availability is a product, not a single number
If the circuit is 99.9% available and your building power is 99.5% available and your router is 99.9% available, the availability of the system your staff actually use is roughly the product of the three: about 99.3%, or five hours a month. Buying a better SLA while running the office router off an unprotected socket is optimising the wrong term. Fixing power and equipment redundancy is frequently cheaper than upgrading the circuit, and it moves the number further.
Questions to put in writing
- What availability, over what measurement period?
- Does degraded performance count as downtime, and against what thresholds?
- What is the committed MTTR, and when does the clock start?
- What are the exclusions, and is scheduled maintenance capped?
- What is the credit schedule, is it automatic, and what is the claim deadline?
- Who measures availability, and can I see the data?
ValoNet states a 99.9% uptime SLA alongside its zero-oversubscription policy. Use the list above on that claim as you would on anyone’s — an SLA you have interrogated is worth more than one you were impressed by, and the same discipline applies when you are deciding between a leased line and business broadband.