Self-optimization of resource allocation in decentralised server farms

Self-supervision and self-adaptation are key requirements for future network and service environment, in which also users' devices play an active role in service delivery. The characteristics of such systems, e.g., distribution, dynamicity, disappearing connectivity, prevent to adopt logically fully centralized solutions for implementing supervision functions. This paper discusses the feasibility of a decentralized approach for optimizing resource allocation in distributed systems, such as server farms structured as clouds of computing resources offered by different administrative domains. The proposed approach relies on local autonomic features, embedded in computing resources, and their cooperation through gossiping on overlay networks. Self-optimization addresses load balancing and power saving issues. Simulations show that, even without centralized control, a global governance of the distributed system emerges from local autonomic decisions and local data exchange.