The Polling Primitive for Computer Networks

We describe a distributed computing primitive termed polling that is both a means of synchronization and communication in distributed or concurrent systems. The polling operation involves the collection of messages from nodes in an interconnection network, in response to a query. We define the semantics of polling, and present algorithms for implementing the operation on complete and hypercube networks. Time and message lower bounds are presented, and it followed by an analysis of the number of operations performed at each node for every algorithm. We show that polling in a complete graph on 2 vertices can be completed in 2n rounds using 2 + 2n−3 + d2n−3+1 3 e − 2 messages. In case of n-cube, we show that polling in 2n rounds requires d2n + 1 32 + 14 √ 2 − 43e messages and we present an algorithm that completes polling in 2n rounds and sends 2 + 3 · 2n−4 − 1 messages.