Collective-adaptive systems offer an interesting notion of interaction where run-time contextual data are the driving force for interaction. The attribute-based interaction has been proposed as a foundational theoretical framework to model CAS interactions. The framework permits a group of partners to interact by considering their run-time properties and their environment. In this paper, we lay the basis for an efficient, correct, and distributed implementation of the attribute-based interaction framework. First, we present three coordination infrastructures for message exchange, then we prove their correctness, and finally we model them in terms of stochastic processes to evaluate their performance.