Egocentric voting algorithms

An important problem in distributed systems is distributed agreement. One form of distributed agreement is approximate agreement (AA) in which non-faulty processes need to agree on values within a predefined tolerance. This paper partitions AA voting algorithms into 3 broad categories: anonymous, egophobic, and egocentric. Each category is further subdivided into families of algorithms. One such family of voting algorithms which belongs to the egocentric category is examined. Ad-hoc analyses of some members of this family of algorithms have been studied individually under an overly conservative fault-model in which all faults are presumed to behave in the worst-case Byzantine manner. This paper develops a methodology to determine quickly the fault tolerance and convergence rate of any member of this family under a hybrid fault model consisting of asymmetric, symmetric and benign faults. The results are weighted against those of several known voting algorithms. A sub-family of egocentric algorithms with optimal performance is identified. Traditionally, egocentric algorithms used the entire voting multiset, as in fault-tolerant mean, to reach a single voted value. It was not known how the distribution of selected elements would affect the convergence rate. Here, convergence is improved considerably if the appropriate largest and smallest data items are not included in the selected multiset