View article

Numerous large-scale decentralized systems assume loosely synchronized clocks. Existing time protocols have not been designed for deployment in such systems, since they are complex and require manual configuration. We present the Gossiping Time Protocol (GTP), a completely self-managing epidemic time synchronization algorithm for peer-to-peer networks. In GTP, each node synchronizes its time by gossiping with other nodes. The decisions regarding sample evaluation and gossiping frequency are purely local, yet they result in consistent behavior of the whole system. Large-scale experimental evaluation of a 64,500-node network emulated on 65 machines indicates high scalability and reasonable accuracy of GTP.