Optimistic Fault Tolerance

Traditionally, fault tolerance has been considered a pivotal component of system design. However, as modern commodity hardware becomes increasingly reliable, building dedicated fault tolerance systems becomes unnecessary. Therefore, while in this paper we propose nothing novel and make no technical contributions, we nevertheless identify a unique approach to fault tolerance, which we dub optimistic fault tolerance. We implement optimistic fault tolerance, or OFT, as a framework implemented as a C library which supports seamless interoperability with any large scale system. We provide a survey of real systems that demonstrates the rarity of system failure, and additionally show through benchmarks and simulation that across numerous workloads, our framework provides tolerance in the face of rare arbitrary failures.