Abstract

This paper indicates that a scalable fault-tolerant name service can be provided utilizing an overlay network and that such a name service can scale along a number of dimensions: it can be sized to support a large number of clients, it can allow large numbers of concurrent lookups onthe same name or sets of names, and it can provide name lookup latencies measured in seconds. Furthermore, it can enable updates to be made pervasively visible in times typically measured in seconds for update rates of up to hundreds per second. We explain how many of these scaling properties for the name service are obtained by reusing some of the same mechanisms that allowed the underlying overlay network to scale. Finally, we observe that the overlay network is sensitive to bandwidth and CPU limitations.