Conquering Complexity

Abstract

It is sometimes argued that with the right training, discipline, and tools it should be possible to produce zero-defect code. Very few things in life, though, are zero-defect—not even things that can be considered life critical. If you practice sky-diving, your main parachute could fail to open, no matter how carefully you check it before each jump. A parachutist would be wise not to trust a company that tries to sell him a zero-defect parachute. The jumper is more likely to avoid problems by bringing a spare chute. That is: the seasoned parachutist takes the possibility of component failure into account in the adoption of a system that has a relatively low probability of system failure. We can provide system reliability, even when none of the system components are zero-defect. In many cases, though, mere redundancy does not solve the problem (i.e., multiple sky-jumpers in parallel). Reliable systems are designed with the possibility of component failure in mind, and with remedies in place to reduce the odds of system failure. Component failure is a rarely and isolated event though. In this chapter we will consider the nature of failure in complex software systems, and how we can develop methods to leverage these insights.