Danzig et all designed an architecture that supports a large group of autonomous servers and provides a way for them to maintain a shared distributed database efficiently. Their architecture uses a flood-d protocol that is based on Golding's TSAE protocol to propagate database updates. Internet Multicast is used to broadcast packets on the Internet.
The key contribution of this paper is the flood-d protocol. Designed to scale to thousands of autonomous hosts, it groups service replicas into multiple, autonomously administered replication groups. This has the effect of reducing the amount of state each replica must keep. Service replicas continuously measure Internet bandwidth and use this information to create optimal update topologies.
The approach used to create these optimal update topologies is attractive because it actually solves for an optimal update topology. They use Steiglitz's algorithm to determine a logical connectivity topology with node connectivity , a minimal diameter, and a minimal edge cost. As long as there is some protection against network failures. They approximate an optimal solution efficiently using a stochastic algorithm known as simulated annealing.
Their simulations show that this approach is efficient. Gathering replicas into groups results in faster update propagation, and taking advantage of the Internet topology in picking neighbors reduces the cost for propagating updates. We expect that using topology information in selecting cache locations will also result in bandwidth and latency savings as well.