Design and analysis of efficient fault-detecting network membership protocols

Network membership protocols determine the present nodes and links in a computer network and, therefore, contribute to making the huge amount of general distributed systems and their inherent redundancy available for fault tolerance. Two different protocols, GNL and GLV, are described in detail which solve the problem for a very general set of assumptions not covered by existing solutions of related research areas like network exploration, system level diagnosis and group membership: neither the topology nor a superset of the nodes are known in advance. Nodes do not require any special initial knowledge except a unique relative signature (Leu, 1994) for authentication. Faults may affect any number of nodes and can cause arbitrary behaviour except for breaking cryptographic fault detection mechanisms like digital signatures. The protocols are compared in terms of execution time and global communication costs based on simulation experiments on randomly generated topologies. It is shown that the GLV protocol dominates the GNL protocol clearly for nearly all practically relevant cases.<
>