[the8472]

http://doc.tm.uka.de/SKademlia_2007.pdf

Combine that with using the global overlay only for bootstrapping of common-interest sub-networks and you'll limit the incentives that an attacker will have to attack the global overlay while also reducing the effectiveness because a single non-fake contact will be sufficient to join the subnetwork.

It's not an absolute defense (sybil-resistance without central authority is hard) but in practice it won't be worth it for attackers.

[ralala]

I think it's still important to find a good balance for the computational cost of the crypto puzzle.

Attackers will probably have a lot of computational resources (think of AWS, botnets, GPU computations), while typical users don't (mobile phones).

I haven't had time yet to look into the IPFS details yet, but thanks for the reference.