When comparing hosting, cloud or colocation providers, the SLA percentage is one of the first figures that appears: 99.9%, 99.95%, 99.99%. It all sounds roughly the same, a tiny difference between numbers already very close to 100%. But that tiny difference has enormous consequences in terms of permitted downtime hours per year, the impact on your business and the infrastructure the provider needs to honour its commitment.
In this article we break down what an SLA is, what the availability "nines" actually mean, what is and is not measured, how downtime affects different industries, what infrastructure is needed to achieve high availability, the difference between a contractual SLA and real availability, and what questions you should ask any provider before signing.
What Is an SLA
An SLA (Service Level Agreement) is a contractual agreement between a service provider and its customer that defines the minimum levels of quality, availability and performance the provider commits to delivering. In the context of hosting and data centres, the SLA primarily establishes the percentage of time the infrastructure will be operational during a given period, typically a month or a year.
An SLA is not just a marketing number: it is a measurable contractual commitment. A good SLA precisely defines what is measured (network, power, hardware), how it is measured (monitoring tools, measurement points), what exclusions exist (scheduled maintenance, force majeure) and what compensation the customer receives if the provider fails to deliver (credits, financial penalties).
The SLA percentage represents the minimum guaranteed availability. An SLA of 99.99% means the provider commits to its service being down no more than 52.6 minutes per year. Any downtime beyond that threshold triggers the compensation clauses.
The "Nines" Explained
The industry uses "nines" to refer to the number of nines after the decimal point in the availability percentage. Two nines (99%) is a basic level; five nines (99.999%) is the standard for the most critical infrastructures in the world. The difference between each level seems insignificant as a percentage, but it is dramatic in terms of permitted downtime:
| SLA | Nines | Downtime/year | Downtime/month |
|---|---|---|---|
| 99% | Two nines | 3.65 days | 7.31 hours |
| 99.9% | Three nines | 8.76 hours | 43.8 minutes |
| 99.95% | Three and a half nines | 4.38 hours | 21.9 minutes |
| 99.99% | Four nines | 52.6 minutes | 4.38 minutes |
| 99.999% | Five nines | 5.26 minutes | 26.3 seconds |
Key fact:
The difference between 99.9% and 99.99% is just 0.09 percentage points, but it represents going from nearly 9 hours of downtime per year to less than 53 minutes. Each additional "nine" reduces permitted downtime by a factor of 10.
What Is Measured in an SLA
Not all SLAs measure the same thing. In a data centre or hosting provider, availability is typically measured across three independent pillars, each with its own SLA:
- lan Network uptime: availability of network connectivity up to the customer's demarcation point. This measures the accessibility of the backbone network and peering/transit links. A network SLA of 99.99% is common in data centres with redundant multi-homed BGP connectivity.
- bolt Power uptime: availability of the electrical supply to the customer's rack or equipment. This includes the feed from the public grid, UPS systems, diesel generators and automatic transfer switches (ATS). A Tier III data centre typically delivers 99.982% power availability.
- dns Server/hardware uptime: availability of the physical or virtual server provided to the customer. This includes hardware (CPU, RAM, disks, RAID controller) and the virtualisation layer if applicable. This SLA is only relevant in managed services or cloud where the provider manages the hardware.
A critical point is the definition of downtime. Most SLAs exclude scheduled maintenance (notified in advance), failures caused by the customer, force majeure (natural disasters, conflicts) and in some cases upstream provider outages. It is essential to read the exclusions before comparing percentages between providers.
Downtime Impact by Industry
The cost of downtime varies enormously by industry. What is a minor inconvenience for a personal blog can represent millions in losses for an e-commerce platform or a financial services company:
| Industry | Primary impact | Recommended SLA |
|---|---|---|
| E-commerce | Direct revenue loss from missed sales, cart abandonment, reputational damage | 99.99%+ |
| Healthcare | Patient safety risk, inaccessible medical records, regulatory non-compliance | 99.99%+ |
| SaaS / Software | User churn, breach of own SLAs with customers, loss of trust | 99.95%+ |
| Finance / Banking | Regulatory risk, failed transactions, fines from supervisory bodies | 99.999% |
| Media / Streaming | Audience loss, contractual penalties with advertisers | 99.95%+ |
How High Availability Is Achieved
Achieving an SLA of 99.99% or higher is not about having a single reliable component, but about eliminating every single point of failure across every layer of the infrastructure. Each critical system must have at least one alternative path that takes over automatically when the primary component fails:
- electrical_services Redundant power (N+1 or 2N): an N+1 configuration means one additional backup component (if you need 2 UPS units, you install 3). A 2N configuration means completely duplicating the entire power chain: two independent utility feeds, two UPS systems, two distribution circuits. Tier III+ data centres use 2N with backup diesel generators.
- ac_unit Redundant cooling: N+1 cooling systems that maintain operating temperature even if one cooling unit fails. This includes continuous temperature and humidity monitoring in every aisle.
- hub Redundant network (multi-homed BGP): multiple transit and peering providers with automatic BGP routing. If one carrier goes down, traffic is automatically redirected through alternative routes with no manual intervention or perceptible downtime.
- storage Hardware redundancy: disks in RAID or distributed storage, dual power supplies, redundant NICs, server clusters with automatic failover. In cloud environments, virtualisation with HA allows VMs to be automatically restarted on another physical node.
- monitoring 24/7 monitoring and fast response: continuous monitoring systems that detect anomalies before they become incidents, with automatic alerts and on-call teams for immediate response.
Fundamental principle:
High availability is not achieved by buying better hardware, but by eliminating single points of failure. Every critical component (power, network, storage, compute) must have an alternative path that works automatically when the primary path fails.
SLA Is Not the Same as Real Availability
It is important to understand that an SLA is a minimum commitment, not a prediction of actual availability. A provider with a 99.95% SLA may have a real availability of 99.999% for years, because the infrastructure is designed with safety margins. The SLA is the floor, not the ceiling.
When a provider breaches its SLA, the compensation clauses are triggered. The most common compensation is a service credit: a percentage of the monthly bill that is deducted from the next billing period. For example, a typical SLA might offer a 5% credit for every 0.1% of downtime below the SLA, up to a maximum of 100% of the monthly bill.
However, SLA credits never compensate for actual business losses. If your e-commerce site loses 50,000 euros in sales during two hours of downtime, a 200 euro credit on your hosting bill does not cover the damage. That is why it is essential to choose a provider whose actual infrastructure (not just its SLA on paper) is designed to minimise the risk of downtime.
SLA in the Cloud vs On-Premise
Cloud providers (AWS, Azure, Google Cloud and European providers like EasyDataHost) offer SLAs per individual service: one for compute, another for storage, another for network. The key nuance is that when your application depends on multiple services, the composite availability is the product of the individual SLAs.
For example, if your application uses compute (SLA 99.99%), storage (SLA 99.99%) and a load balancer (SLA 99.99%), the composite availability is 99.99% x 99.99% x 99.99% = 99.97%. Each additional service in the chain reduces total availability. This means simpler architectures with fewer dependencies have an inherent advantage in availability.
In colocation environments, the data centre SLA covers power, network and the physical environment, but the customer is responsible for the availability of their own hardware and software. This offers more control but also more responsibility. With managed services, the provider assumes full responsibility for the stack, simplifying SLA management for the customer.
Questions to Ask Your Provider
Not all SLAs are equal, and a high percentage means nothing if the exclusions are too broad or the compensation insignificant. Before signing with any provider, ask these questions:
- help What exactly is measured: network, power, hardware, software, everything? An SLA that only covers the network but not power leaves critical gaps.
- help How it is measured: what monitoring tool is used, from which points measurements are taken, how downtime is calculated. If only the provider measures it with no transparency, there is a conflict of interest.
- help What exclusions exist: scheduled maintenance, force majeure, third-party failures, DDoS attacks. Some exclusions are reasonable; others can hollow out the SLA entirely.
- help What compensation you receive: service credits, percentage per point of non-compliance, maximum cumulative amount. If the maximum credit is 10% of the bill, the real penalty for the provider is minimal.
- help What is the real track record: ask for historical availability data for the last 12 months. A serious provider will publish its uptime statistics and incident log.
SLA at EasyDataHost
EasyDataHost offers an availability SLA of 99.95%+ backed by infrastructure designed to exceed that commitment. Our data centre in Madrid operates with redundancy at every critical layer: 2N power supply with UPS and diesel generators, multi-homed BGP network with multiple carriers, distributed storage with triple replication and 24/7 monitoring with a response team.
The real availability of our infrastructure consistently exceeds 99.99%, because the architecture is designed with safety margins that go beyond the contractual SLA. Every incident is documented, root cause is analysed and preventive measures are implemented to avoid recurrence.
- check_circle Contractual SLA 99.95%+: documented commitment in the service level agreement with clear compensation clauses.
- check_circle Redundant infrastructure: 2N power, multi-homed network, distributed storage at the Madrid data centre.
- check_circle 24/7 monitoring: proactive anomaly detection with automatic alerts and a response team.
- check_circle Transparency: historical availability data and incident log available to customers.
Conclusion
An SLA is not just a percentage on a website: it is a contractual commitment that reflects the quality of a provider's infrastructure, processes and investment in reliability. Understanding what the "nines" mean enables you to evaluate whether the availability level on offer is adequate for your business needs.
- arrow_right An SLA is a measurable contractual commitment of minimum availability.
- arrow_right Each additional "nine" reduces permitted downtime by a factor of 10.
- arrow_right High availability is achieved by eliminating single points of failure at every layer.
- arrow_right The contractual SLA is the guaranteed minimum; real availability is usually higher.
- arrow_right EasyDataHost offers a 99.95%+ SLA with redundant infrastructure and 24/7 monitoring.
If you need infrastructure with guaranteed high availability for your project, contact our team to design the solution that best fits your availability requirements.