Matthew Toseland wrote: > Hmmm. Sounds unscalable... Is it global horizons? Does it scale > sublinearly?
The horizon is fixed and small, so it would be suitable for a darknet. They don't give details of the local routing protocol but any proactive ad hoc protocol would do. The message overhead for inserts/requests is O(sqrt(n)), but the delay is only O(log(n)^k) for a wide range of graphs because it uses parallel inserts/requests. On the other hand you can achieve O(sqrt(n)) message overhead simply by floooding every insert and request to the nearest sqrt(n) nodes, and the expected delay will also be O(log(n)) for a wide range of graphs (including regular random graphs, scale-free graphs and Kleinberg small worlds), so I'm not sure how impressed I am... Cheers, Michael
