Fault - Containment in Self - Stabilizing Distributed Systems

Self-stabilizing systems can automatically recover from arbitrary transient faults, and changes in the environment of the system, without any external intervention. However, in existing distributed self-stabilizing protocols, the performance of recovery is not linked to the severity of the fault. Recovery from failure at even a single component of the system may take a long time and aaect the operation of the entire system. Since at any given time, limited faults in a small number of components are more likely than faults in a large number of components, this limitation restricts the use of self-stabilizing protocols in practice. As a solution, we propose in this thesis the design of fault-containing self-stabilizing protocols, self-stabilizing protocols that provide additional performance guarantees from less severe faults. From limited faults that are expected to occur more frequently in practice, such protocols ensure very fast recovery, and aaects only the processes in a small region around the faults during the recovery. The eeects of limited faults are thus contained eeciently in both time and space. At the same time, self-stabilization ensures automatic recovery from occasional transient faults that may be more widespread in the system. These protocols are a step forward towards making self-stabilization more practical. As a rst step, we focus on fault-containment from single transient faults. However , our deenitions are applicable to containment from other subsets of limited faults as well. The rst part of this thesis presents a framework for deening and evaluating fault-containing self-stabilizing protocols. In particular, we deene important metrics 2 for evaluating the fault-containment properties of a self-stabilizing protocol. The second part of the thesis presents fault-containing self-stabilizing protocols for three important problems: leader election on a ring, construction of the spanning tree of a network, and construction of the breadth-rst-search tree of a network. Finally, we present a general technique to automatically transform a non-reactive self-stabilizing protocol into a fault-containing self-stabilizing protocol. This general technique sig-niicantly simpliies the design of fault-containing self-stabilizing protocols. Abstract approved: Thesis supervisor Title and department Date Thesis supervisor Title and department