Design for failure: Software challenges of digital ecosystems

Abstract

In dynamic computation ecosystems involving many different participants, a topdown approach to system dependability does not work. It is not possible to take a topdown approach to system design and implementation and to validate the resulting system against some specification. Rather, we have to assume that elements of the system will become unavailable at unpredictable times and that some elements may be unreliable. Instead of designing systems to avoid failure, we must re-orient our thinking and design systems so that we can tolerate failure and recover from failures when they occur. In this talk, I will discuss the challenges of designing for failure and will introduce research on responsibility modelling that provides information for failure recovery.