Hierarchical decision structure for distributed algorithms

When the number of nodes in a distributed system becomes very large, neither a fully centralized algorithm nor a fully distributed algorithm is suitable for implementation. The number of messages exchanged in most distributed algorithms becomes too large to make them practical, while the system with a fully centralized algorithm is vulnerable to failure of the control node. The authors propose another approach to constructing an algorithm for distributed systems with a large number of nodes, in which the number of messages to be exchanged would be reduced at a cost of communication delay if needed, and fault-tolerant capability remains. The nodes in a distributed system are partitioned into multiple groups, each of which contain multiple nodes. The process of making a decision is also divided into two steps: a decision in a group is made first, followed by a decision made based on the group decisions. One can apply different algorithms at each decision level. The authors examine the applicability of the approach using well-known solutions to two problems: mutual exclusion and majority consensus. As a result, they believe that this kind of framework provides another way of designing algorithms for distributed systems