This paper addresses decentralised workflow scheduling, which calls for fulfilling two seemingly contradictory requirements: decentralisation and efficiency. We describe a two-layer architecture that allows decentralisation while making it possible for each scheduling decision to be taken based on a global perspective of the current state of resources. The first layer expresses the scheduling strategy on a global perspective, relying on a coordination space where workflows are first decomposed in tasks, and then tasks mapped onto resources. The second layer allows this global policy to be enacted in a fully-decentralised manner, based on a distributed hash table indexing resources, enhanced with advanced discovery mechanisms. Thus, in spite of decentralisation, the system is able to select the momentarily most appropriate resource for a given task, independently of the location of the provider of the resource. The framework’s expectations in terms of scalability and network overhead are studied through simulation experiments.