Byzantine Processors and Cuckoo Birds: Confining Maliciousness to the Outset

TL;DR

This paper demonstrates that Byzantine malicious behavior can be confined to a simple, undetectable egg-replacement model similar to cuckoo birds, allowing Byzantine problems to be reduced to benign failure models and unifying distributed computing theories.

Contribution

It establishes a correspondence between Byzantine and benign failures, enabling the reduction of Byzantine computability to benign failure analysis and introduces Recoverable Reliable Broadcast.

Findings

Byzantine maliciousness can be modeled as undetectable input replacements.

Results about correlated faults in benign settings apply to Byzantine models.

Processors can output once faults cease, similar to benign failure models.

Abstract

Are there Byzantine Animals? A Fooling Behavior is exhibited by the Cuckoo bird. It sneakily replaces some of the eggs of other species with its own. Lest the Cuckoo extinct itself by destroying its host, it self-limits its power: It does not replace too large a fraction of the eggs. Here, we show that any Byzantine Behavior that does not destroy the system it attacks, i.e. allows the system to solve an easy task like epsilon-agreement, then its maliciousness can be confined to be the exact replica of the Cuckoo bird behavior: Undetectably replace an input of a processor and let the processor behave correctly thereafter with respect to the new input. In doing so we reduce the study of Byzantine behavior to fail-stop (benign) behavior with the Cuckoo caveat of a fraction of the inputs replaced. We establish a complete correspondence between the Byzantine and the Benign, modulo different thresholds, and replaced inputs. This work is yet another step in a line of work unifying seemingly distinct distributed system models, dispelling the Myth that Distributed Computing is a plethora of distinct isolated models, each requiring its specialized tools and ideas in order to determine solvability of tasks. Thus, hereafter, Byzantine Computability questions can be reduced to questions in the benign failure setting. We also show that the known results about correlated faults in the asynchronous benign setting can be imported verbatim to the asynchronous Byzantine setting. Finally, as in the benign case in which we have the property that a processor can output once its faulty behavior stops for long enough, we show this can be done in a similar manner in the Byzantine case. This necessitated the generalization of Reliable Broadcast to what we term Recoverable Reliable Broadcast.