[General] Concept of P2P DHT

oranges email at oranges.net.nz
Thu Feb 4 14:53:26 PST 2016


Here is my high level understanding.

There is a distance function that defines for any given node, how
close a set of keys are to it (the closeness is arbitrary, it just
needs to be a consistent algorithm)

If your node is "close" to a key you store those key -> ipaddress
mappings on your machine. (To make this storage more robust, the key
-> ipadress mappings are a usually replicated on N machines, where N
is some arbitrary value of how resilient the DHT is.)

Now, you can't obviously know about all the nodes in a network. So
usually, you only know K nodes that are spread in either direction of
your "closeness value", i.e some will be closer to the zero, than you
and some will be closer to the infinity. (ideally half and half).
Although in this case I'm not sure exactly how tox chooses candidates
for you to know about (it may just be, the last K nodes I have spoken too)

To lookup a key, you find the 3 nodes that you know about that are
closest to the key that you want to find.

They may not actually know about the key, but what they can do is then
find the 3 closest nodes to the key that they know about and ask them.

This repeats and you eventually converge closer and closer to the key
you want until you find a node with the key -> ipaddress mapping and
it is returned to the original asker.

This is why when you first time you ever connect to the DHT you need
to use one of the bootstrap nodes at nodes.tox.chat, so that you can
"join" the DHT and become aware of at least some members of the network.

(Note that this simplifies the real situation as in tox.chat's case,
there is onion routing involved, so that you can't just trivially get
someones IP address by friend requesting them (they have to accept the
request)

Hopefully that clarifies a bit how the DHT works


More information about the General mailing list