Bell Labs, USA
Invited Lecture: Toward survivable systems
Abstract : As the world moves toward increasing reliance on distributed computer networks, it is essential to find ways to build such infrastructures that perform reliably under a wide range of circumstances that may include both accidents and malicious attacks. A "survivable" system is one that can make meaningful progress even when some (human or computer) components fail to behave as expected, and particularly when they behave in a way as to undermine the correct operation of the system as a whole. In this talk we survey our latest progress in a five-year (and counting) research program on constructing survivable systems. Our discussion will focus on our current development of a survivable, distributed object store called Fleet, and the technical challenges in scale, intrusion detection, and resilient update propagation that our design addresses. We will also discuss challenging applications that have shaped this design, including an electronic voting system for Costa Rica's national elections.