Skip to content

What is an SLA?

A service level agreement (SLA) is the part of a contract that turns "we'll keep your systems running" into a number: which service, measured how, kept above what threshold — or else. It's the difference between a promise and a commitment: a promise has good intentions, a commitment has consequences.

What an SLA actually contains

A usable SLA answers five questions:

  • Scope — which systems are covered. "The e-commerce site and the mail server", not "IT".
  • Metric and target — what is measured and the threshold: for example, 99.5% monthly uptime.
  • Measurement window — the period the percentage is computed over, usually the calendar month.
  • Exclusions — what doesn't count: agreed maintenance windows, failures outside the provider's control (the client's own ISP, for instance).
  • Remedy — what happens on a miss: typically service credits.

Targets and measurement windows

The window changes the meaning of the same number. 99.9% measured over a year permits one 8.8-hour outage in December and still shows a clean annual figure; 99.9% measured monthly caps every single month at about 43 minutes (the nines of availability). Monthly windows are the norm in managed-services contracts: they reset, they match billing, and they keep one bad week from being buried under eleven good months.

Pick the target honestly — the architecture has to support the number, which for a typical SME workload usually lands at 99.5%–99.9% monthly (what is downtime covers what eats the budget).

Why an SLA needs continuous, independent measurement

An unmeasured SLA is decoration. Without data, a missed target surfaces as an argument: the client remembers "it was down all morning", the provider remembers "a brief interruption", and nothing settles it. Two properties fix this:

  • Continuous — checks every minute, all month. The uptime percentage becomes a computation over thousands of timestamped results, not an estimate.
  • Independent — measured from outside the system being judged. A server can't be trusted to report its own death; when it goes down, its logs and dashboards often go with it.

With both, the monthly number is the same for everyone, and disputes turn into lookups.

SLA targets in UpAll

In UpAll, every client can carry an SLA target. Checks run every minute against the client's monitored systems, and on the 1st of each month — on paid plans — a PDF report per client compares measured uptime against that target, with downtime minutes, the incident log, p95 latency, and measurement coverage (monthly SLA reports, what's in a report). Reports land in the console's archive; you download and forward them yourself, which keeps you in control of the conversation around the numbers.

An SLA is one of a trio of related terms — the others are the internal objective and the raw measurement. See SLA, SLO, and SLI.