Falling Off the Cliff: When Systems Go Nonlinear

As the systems we build become more complex, understanding and managing their behavior becomes more challenging. If the system’s inputs are within an acceptable range, it will behave predictably. However, the system may “fall off the cliff” if input values are outside this range. This nonlinear behavior is undesirable, because the system no longer behaves predictably: it may not be possible to use, control or even recover the system. In this paper, we describe what it means for a system to fall off the cliff. We outline methods for detecting and predicting these modes of nonlinear behavior, and propose several approaches for designing systems to cope with these instabilities, or to avoid them altogether. We conclude by outlining open research questions for investigation by the systems community. 1 Author affiliations and email addresses: University of Victoria, [email protected]; MIT, [email protected]; Microsoft Research, [email protected]; Rice University, [email protected]; IBM Research, [email protected]; AT&T Research, [email protected]; HP Labs, [email protected]; Rutgers University, [email protected]; Vanu, Inc., [email protected]; Harvard University, [email protected]; University of Cambridge, [email protected]