Robust architectures for open distributed systems and topological self-stabilization

Abstract

Distributed systems are often dynamic in the sense that there are frequent membership changes (nodes joining and leaving the network), either due to regular churn or due to an attack. Maintaining availability and full functionality of such a system under continuous topological changes hence constitutes an important algorithmic challenge. This paper reports on some of our recent results on robust distributed systems. We review two randomized architectures that build upon the continuous-discrete approach by Naor and Wieder, namely the SHELL network which allows for fast joins and leaves and organizes more reliable (or stronger) nodes in a core network where their communication is not affected by malicious (or weak) nodes, and the Chameleon network whose replica placement strategy and whose intentional topological updates ensure resiliency against denial-of-service attacks, even from past insiders. To complement our investigations on randomized architectures, we discuss algorithms to maintain hypercubic networks under worstcase churn. Finally, we advocate the design of self-stabilizing topologies-a very appealing and still not well-understood notion of robustness-that converge quickly to a desirable structure from arbitrarily degenerated states. As a use case, graph linearization is examined in more detail. This invited paper complements the WRAS'10 talk and is joint work with

Stability in Software Engineering: Survey of the State-of-the-Art and Research Directions

“…Stability is evaluated using the worst case of output changes when the input changes at most once for each processor in the system [70]. The study of Schmid [74] focused on structural (topological) self-stabilisation of distributed systems, to allow dynamic convergence to the desired structure after performance deterioration and ensure continuous availability and functionality.…”