Partitioning for efficient consensus
Michael Barborak, Miroslaw Malek · 2002
Consensus protocols are expensive in terms of the number of messages, and ultimately the time, required to execute globally for a large number of processors. The hierarchical partitioning method (HPM) proposed decreases the number of messages and time required to reach consensus by partitioning the set of processors, and organizing these partitions in a hierarchical manner. The authors describe the HPM and present two algorithms, one based on system-level diagnosis and another on Byzantine agreement, that implement the HPM. The algorithms are analyzed in terms of their costs and their impact on the system and are shown to be a large improvement over single-level consensus algorithms. As shown by the analysis, the HPM is a cost-effective method for reaching consensus in large, distributed systems.>