Package freenet.node.probe

Provides network probes which query a node selected - ideally - uniformly at random from the network for a result type requested by the probe originator.

See: Description

Interface Summary

Interface	Description
Listener	Listener for the different types of probe results.

Class Summary

Class	Description
Probe	Handles starting, routing, and responding to Metropolis-Hastings corrected probes.

Enum Summary

Enum	Description
Error
Type	Describes the result type requested by the originating node.

Package freenet.node.probe Description

Provides network probes which query a node selected - ideally - uniformly at random from the network for a result type requested by the probe originator. The Metropolis-Hastings algorithm is used for node selection. After selecting a random peer, the node has a localDegree / remoteDegree probability of accepting a forward to that peer. This is in contrast to a uniform walk, where a randomly selected peer is forwarded to with no further consideration.

Because nodes with high degree by definition have more connections, there are more chances to forward to them, and a random walk will over-represent them. Metropolis-Hastings correction corrects for this bias. To give an intuition for why this works, an example:

A network has four nodes. One central node is connected to the three others. The others are connected only to the central node. With uniform routing, from the central node it can go to any of the three others, and from a leaf it always returns to the center. Half of the time the probe is on the center node, and the other half of the time is split between the three leaf nodes. It spends 1/2 of the time on the center and 1/6 on each leaf.

With Metropolis-Hastings routing, the central node always (3/1) forwards to the leaf nodes, but the leaf nodes only forward to the central node 1/3 of the time. The rest of the time the probe remains on the leaf and HTL decrements. The probe will spend 1 hop on the central node and, on average, 3 on a leaf. Even though the central node only spends 1 hop at a time, it still connects the others, so its single hops add up to equal those spent on each leaf node. Provided sufficient initial HTL to both allow travel across the network and for the probability distributions to run enough times to even out, an endpoint reached with Metropolis-Hastings correction is chosen uniformly at random from the entire network.