Region-Adherent Algorithms: Restricting the Impact of Faults on Service Quality

Abstract

We present a new class of fault-tolerant distributed algorithms based on a concept which we call region adherence. A region-adherent algorithm upper-bounds the violation of safety due to faults in space. Region adherence counter-poses the concept of self-stabilization which upper-bounds a violation of safety in time. It turns out that region adherence is an orthogonal concept to self-stabilization. We give a formal definition of region-adherence that, intuitively, upper-bounds the reduction of the algorithm's service quality per fault. Then, we present a sample algorithm that exhibits region-adherent behavior and prove this property formally. Finally, we analyze the service quality of the sample algorithm via simulation and compare it to the worst-case behavior stated by the region adherence property.