SLA

SLA (Service Layer Agreement) for large distributed software is very important. It plays the role of contract between the application and it's clients. It's also not very straightforward to achieve, especially if many components participate in the process. Each component in this case should have even more strict SLA, because the result of each component SLA would sum up. For example, if single call to some service A is resulted in multiple calls to other services, it's important that other services had better SLA in order to keep service A SLA promise.

There might be different types of SLA: resources, availability, durability and performance.
Many systems provides contract on how many resources are available for the client: memory, CPU, disk space etc.
Some websites and services say that their availability SLA is 99.9%, which means that 99.9% of time they going to be available. Actually, that is not very much at all. There is a nice table on Wikipedia with conversion between availability percentage and actual time.

Some services, especially the storage services like S3, have also durabitlity SLA. This to say how rarely the service might lost the data.

Performance SLA is common for running services that need to not only be available, but return response on request within specified period of time. For performance SLA, it is always common to use some percentile of requests that would be handled within SLA. For instance, it might be said that SLA is return response in 10 ms or less for 99.9% of requests.