Practical distributed systems are typically built under a
crash-fault model. Recently, arbitrary faults such as bit
flips have been observed surprisingly often [5], and have
disrupted large services such as Amazon S3 [2].
We present a framework for building distributed pro-
tocols which automatically improves their fault coverage
by means of an encoded processing compiler [4]. Al-
though encoded protocols cannot withstand attacks by
malicious adversaries, they can tolerate a wide variety
of non-malicious arbitrary faults.
This preliminary work focuses on leader election, a
fundamental primitive in distributed systems. In these
protocols, a bit flips can possibly violate liveness and
safety properties, for instance, by never electing a leader,
or by electing more than one leader at the same time. We
implement two election algorithms in our framework and
experimentally analyze the transformation’s overhead on
CPU utilization and election time