Embracing Failure: A Case for Recovery-Oriented Computing (ROC)

Motivated by the lack of availability demonstrated by current approaches to building servers for the Internet environment, we argue for a new approach to building highly-available systems that better reflects the realities of the modern server environment, namely that failures of hardware, software, and humans are inevitable. Our approach, denoted recovery-oriented computing (ROC), recognizes the inevitability of unanticipated failure and thus emphasizes recovery and repair rather than simple fault-tolerance. We define the properties that a ROC system must provide, and briefly consider how they might be achieved.