Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article6 min readViews0

SLA 99.9% for Hosting: How Much Downtime Does That Percentage Hide?

Converting 99.9%, 99.95%, and 99.99% into minutes, we examine service coverage, exclusions, and service credits to compare hosting plans by their terms rather than by the number of nines.

Comments 0

Blue contract, orange stopwatch, and the server as an SLA metaphor for hosting
In this article

A 99.9% availability SLA for a month allows for approximately 43 minutes and 12 seconds of downtime in a 30-day period, provided the contract actually counts every such minute and applies the metric to your configuration. For a 31-day month, this allowance rises to 44 minutes and 38 seconds. Therefore, before choosing a hosting provider, verify not only the percentage but also the service coverage, the definition of unavailability, exclusions, the measurement method, and the procedure for obtaining a service credit. A high percentage alone does not guarantee that your online store will never go offline.

Convert the percentage to minutes

The monthly figure is calculated using a simple formula: allowable downtime = number of minutes in the billing month × (1 − SLA / 100). Since months vary in length, a 'per month' value cannot be automatically converted into a single fixed annual limit.

Advertised availability

30 days, 43,200 minutes

31 days, 44,640 minutes

99.5%

3 h 36 min

3 h 43 min 12 s

99.9%

43 min 12 s

44 min 38 s

99.95%

21 min 36 s

22 min 19 s

99.99%

4 min 19 s

4 min 28 s

Calculation example: a 30-day month contains 43,200 minutes. If confirmed downtime totals 62 minutes, the observed availability using the simple formula is (43,200 − 62) / 43,200 × 100% ≈ 99.856%. This is below 99.9%. However, drawing a contractual conclusion is premature: first, you must apply the specific definition of downtime and the provider's exclusions. This is a calculation, not a measurement of a specific service.

Check which architecture the percentage refers to

Here is the interesting part: different configurations from the same provider can have different availability targets. In the current AWS agreement for Amazon EC2, the regional 99.99% figure applies when all running instances are simultaneously distributed across two or more availability zones within a single region; if a region has only one zone, the requirement is at least two regions. For a single instance, the specified figure is 99.5%. In the current SLA for Google Compute Engine for the Premium Tier, regions other than Stockholm receive 99.99% for multiple zones, 99.9% for a single instance of most families, and 99.95% for a single instance of the memory-optimized family. For Stockholm, the multiple-zone figure is 99.95%, and the single-instance figure for other families is 99.9%. These examples do not compare cloud quality; they illustrate why a number without context regarding placement scheme and region explains very little.

  • the exact name of the service, tariff, region, and network level;

  • one server or several independent instances;

  • whether distribution across zones, a load balancer, or another fault-tolerance scheme is required;

  • whether the figure is calculated per instance, project, region, or the entire account.

On the Yandex Compute Cloud service page, the SLA for virtual machine availability is listed as 99.95%, while the documentation links to separate general terms and service levels. This is a good reason to open not only the product card but also the related contract document: the short figure on the display does not reveal the calculation method.

Clarify what counts as downtime

A user may see a non-functional shopping cart while the virtual machine responds to network requests. Conversely, external monitoring may record a network error from a single point even though the service is accessible via another route. SLAs typically describe a specific technical event: for example, loss of external connectivity, disk unavailability, or simultaneous unavailability of all provisioned instances.

  • which user or infrastructure trigger starts the countdown;

  • what minimum event duration is taken into account;

  • whose logs and timestamps are considered the time source;

  • whether scheduled maintenance, customer errors, third-party services, and quota overruns are included;

  • how multiple short events are combined into a single downtime period.

The public status page is useful for observation but does not replace a contract. The General iT status page explicitly states that percentages are based on completed checks from a specific point and do not constitute a claimed SLA. In the public offer, availability metrics or compensation apply only under specific agreement. For your service, the order and applicable terms take precedence over the number shown on the general page.

Service credits do not equal compensation for business losses

In cloud SLAs, the typical result of a breach is a credit toward future payments for the affected service, not an automatic payout for lost order value. AWS requires submitting a claim with specified data by the end of the second billing period following the incident. Google requires contacting support within 60 days of the right to a service credit arising, along with logs containing dates and times of unavailability periods. For another provider, deadlines, evidence requirements, and credit amounts may differ.

If a monthly service in the example costs 12,000 ₽ and the terms offer a 10% credit, the result is 1,200 ₽ for future payments—only if the application rules are met. This is not an assessment of store damage nor a promise from a specific provider.

Six questions before payment

  1. Where is the SLA defined: in the tariff, order, contract, or a separate appendix?

  2. Which specific service and configuration are covered by the metric?

  3. How are the start and end of downtime defined, and what events are excluded?

  4. Over what period is the percentage calculated: calendar month, billing month, or another interval?

  5. What service credit is provided, to which amount does it apply, and is an application required?

  6. What timelines, logs, ticket numbers, and other evidence will be required?

If answers remain only in promotional descriptions, request the supplier's exact terms before placing an order. When comparing standard virtual hosting, separately evaluate resources, limits, support, backups, initial and renewal pricing; an SLA addresses only one part of the decision.

What to preserve after an incident

  1. start and end times with time zone;

  2. affected domain, server, region, and user scenario;

  3. HTTP codes, network errors, and logs without secrets or personal data;

  4. results from independent monitoring and provider reports for the same interval;

  5. ticket number and support response;

  6. percentage calculation using the formula and definitions specific to your SLA.

A sufficient selection result is not the maximum number of nines, but a clear combination: a covered configuration, a measurable unavailability indicator, known exceptions, a working method for recording, and a real process for obtaining a credit. Sources and terms were verified on November 3, 2026. Before purchasing or applying for compensation, check the current version of the contract: SLAs change, and the observed availability of a specific site also depends on the application, DNS, database, integrations, and your architecture.

Discussion 0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.