SLA stands for Service Level Agreement.
A service-level agreement is a contract signed between a customer and a provider which outlines the minimum standards the service must meet as guaranteed by the provider.
Performance can be measured in several ways but the most universal is availability, also called uptime. Uptime measures the amount of time that a service is able to be used by its customers (the service is “up” or available, not “down”).
To take electricity provision as an example, your electricity company could promise to supply you with power for a minimum 350 days every year.
Uptime is often measured as a percentage so 350 days in a 365-day year would be 95.89%. Many cloud-service providers commit to 99.9% uptime, which in a year is more than 364 days of uptime in a year. In fact, it is only 8.76 hours a year, which can be broken down into 43.2 minutes a month or 10.1 minutes a week.
Uptime of 99.9% is sometimes referred to as “three nines” uptime and is considered to be high availability. Higher levels of availability include “five nines”, or 99.999% uptime. SLAs guaranteeing greater availability will cost more.
Availability can also be measured as a probability, such as an anti-spam service with an SLA that 0.0003% of emails will be wrongly identified as spam. Apart from uptime and spam effectiveness, SLAs can measure statistics such as latency, fault response and virus detection.
SLAs only have teeth if they are backed up by a financial penalty for failure to provide the agreed level of service. The most common approach is to waive fees for the service for a period determined by the size of the gap between promised and actual availability. Again, the details are worth checking with your vendor.
Trend Micro lists several questions worth asking vendors.
- Does the SLA cover all critical service components including availability, spam-blocking effectiveness, virus infection, false-positive rates, support responsiveness, and email delivery latency?
- Does the SLA provide aggressive remedies for downtime in a given month? Does it measure downtime in minutes or hours?
- Does the SLA provide sufficient compensation for failing to meet commitments? For example, does 1% downtime in a given month – almost one full working day – provide a credit for only 1% of the monthly charge?
Have a look at these examples of SLAs for major cloud providers.