How to Calculate System Availability and Reach More ‘9’s in Your SLA

When evaluating a system's availability and reliability, we often refer to "three nines" or "four nines" to express the Service Level Agreement (SLA) and the expected downtime per year.

This article translates and expands on an English source, covering how those nines are calculated, the factors to consider, and practical methods to achieve higher availability.

1. System Availability

System availability is modeled by arranging components in series or parallel configurations.

If a component's failure makes the whole system inoperable, the components are considered to be in series.

If a component's failure can be taken over by another component, the components are considered to be in parallel.

1.1 Serial Availability

When two components X and Y are in series, the system is available only if both X and Y are simultaneously available; the overall availability is the product of the two components:

Thus, a series system’s overall availability is always lower than that of any individual component.

Example calculation for components X and Y:

1.2 Parallel Availability

When two components are in parallel, the system remains operational as long as at least one component works. The overall availability is:

Consequently, a parallel system’s availability is higher than that of any single component.

2. Availability Calculation Example

2.1 Understanding the System

The system consists of input sensors, two redundant signal processors (primary and standby), and output converters. The standby processor monitors the health of the primary processor.

2.2 System Reliability Model

The hardware and software of each processor are modeled as separate entities; they are in series because both must work for the processor to function.

The two processors (hardware + software) form a parallel group, allowing the system to continue operating if one processor fails.

The input sensor, processor group, and output sensor are placed in series, so failure of any one causes total system failure.

2.3 Calculating Component Availability

Component availability is derived from MTBF (Mean Time Between Failures) and MTTR (Mean Time To Repair):

For hardware we estimate MTBF from vendor datasheets and assume an MTTR of about 2 hours. Software MTBF is approximated as the time between restarts (≈ 4000 hours) and MTTR as the restart time (≈ 5 minutes), which includes:

Lost time due to software crashes.

Detection time of the failure.

Time to restart and return to service.

Key observations:

Even with higher hardware MTBF, software often shows higher availability because its MTTR is much lower.

Input and output sensors have high availability, contributing positively even without redundancy.

2.4 Calculating System Availability

The final system availability is computed by applying the series and parallel formulas to the component values.

3. How to Achieve More ‘9’s

Different organizations define the required number of nines differently; many internet companies target 99.99 % (four nines), while some public‑service sites may only aim for 99.9 %.

Lower availability leads to greater loss, especially during critical moments when a single minute of downtime can cost a large order. Therefore, maximizing SLA availability directly improves business productivity.

To reach higher nines, continuously monitor services, respond quickly to incidents, and add redundancy to eliminate single points of failure.