Detail

Fault tolerance: there is no B

In this talk I will discuss a different approach to discuss fault tolerant systems, one which eschews the traditional distinction between crash and Byzantine failures and instead focuses on the development of systems that are simply fault tolerant, for any environment. While systems designed in this manner remain optimal with respect to replication requirements, performance optimizations may be applicable to specific failure configurations. The two keys to making this possible are (a) making a strong distinction between safety and liveness and (b) recognizing that commission failures exist outside of the traditional failure hierarchy.