Gossip-Based Dissemination of Time

Time synchronization between computers within a very large, highly dynamic network is a challenging task. Current solutions operate mostly in a hierarchical client-server mode based on a static configuration of logical connections, which tends to lack the scalability and robustness to failures. \n \nIn this thesis, the author presents the Gossiping Time Protocol (GTP) --- an approach to time synchronization employing the theory of epidemics. GTP is a completely decentralized solution in which all the hosts form a peer-to-peer network. They gossip with each other in order to propagate accurate time. The algorithms constituting GTP have desired properties of scalability and robustness, while offering fast and quite accurate synchronization. Experimental results obtained with a prototype implementation on an emulated network of more than 64,000 hosts scattered across the machines of a wide-area cluster computer confirm the above claim.