Abstract
Peer-to-Peer (P2P) networks have become prevalent recently, thanks in large part to the publicity surrounding file-sharing networks, but P2P is evolving to encompass a wide- ranging set of applications. For many of these, a resource discovery mechanism is an essential basic service, but the properties of P2P networks make provision of this a nontrivial task. Solutions proposed have included central indexes, flooding and message forwarding, but the most promising appears to be the use of Distributed Hash Tables (DHTs). DHTs have been used to provide data lookup within logarithmic message costs whilst only requiring maintenance of limited amounts of routing state. One of the most widely known DHTs is Chord, which provides lookup in typically O(log n) hops across the network, where n is the number of nodes in the structure. Understanding that this message cost is proportional to the network size, our contribution is ROME (Reactive Overlay Monitoring and Expansion), a set of processes which run on top of the Chord DHT to provide control over network size. Every node acts as an autonomous agent in order to react to node underload and overload events, trying to maximise capacity utilisation and only increase the size of the ring when existing capacity is inadequate. Through simulation and calculation we show ROME can reduce the hop counts in networks where available node capacity exceeds workload, closely converging with that of Chord where capacity and workload become equal.