Like people, real systems suffer from the vicissitudes of a whimsical reality. An architecture built according to the basic precepts discussed thus far would doubtless be flexible, scaleable, and able to handle the desired business functions with panache—but when actually deployed, it would probably fail in some way. Like people, real systems suffer. Your system might freeze at 12:17 AM every Saturday, slow to a crawl during a critical usage period, crash only when the project sponsor's boss logs in, or simply be hit by a power failure during a critical transaction. Above and beyond the dictates of good theoretical system design, systems architectures in the real world must be designed (and constructed) with an eye for the uncertain world of deployment.

Required Characteristics

Distributed systems must possess certain characteristics to function reliably after deployment. Several of these characteristics pertain to maintenance of quality of service, which is essential in the face of unforeseen interruptions in the flow of operations.

• Stability. When one part of a system fails, it can cause other parts of the system to fail. Even if a failed subsystem restarts, it may not be able to initialize itself properly if it depends upon other system elements that have also failed. Thus, the system may thrash as the failure cascades through the system in a chaotic manner. Dependable systems must be stable.

• […]