On Thu, Aug 14, 2008 at 11:01 AM, Salman Abdul Baset <[EMAIL PROTECTED]> wrote: > > Nodes need to connect to a central server to get their identity signed by a > central server. This must happen to prevent Sybil attacks.
Just wanted to comment on this commonly-made assertion: central-server authentication does not prevent a Sybil attack, but only makes it harder for someone with limited resources to mount (e.g., it is fairly easy to detect a naive Sybil attack originating from a single IP address or a particular domain if you have a central authority). But a central server can certainly be tricked into issuing multiple authentications for multiple identities which all end up belonging to a single entity. The central server does not solve this problem, unless you strongly tie identities to something in meatspace such as a credit card number or government ID/government clearinghouse. And even then, it only prevents the Sybil attack weakly (an entity with sufficient resources can still mount the attack). Imagine a Sybil attack mounted by the controllers of the Storm Botnet, for example. Centrally signing identities with crypto alone in this case buys you virtually nothing over self-certifying identifiers and requiring nodes to offer proof of resources. Even if you throw in something like a 1-credit-card-per-id requirement, you'd have to face the fact that your adversary may have access to thousands of credit cards. I think the belief that a central authority solves these problems stems from the original, and quite excellent "Sybil Attack" paper (http://www.cs.rice.edu/Conferences/IPTPS02/101.pdf), in which Douceur shows that distributed authentication schemes alone are provably insufficient to solve this problem. The language in the introduction alludes to the idea that central authorities are the answer, but this is never asserted outright, and is certainly not proven. In fairness, if you limit the scope of the Sybil attack problem to "prevent a single node from generating too many identities," or if you use the term "entity" interchangeably with the term "computing node" then some of my argument goes away, and that may be what was originally intended. But I think the semantic difference is important, especially in the presence of well-heeled adversaries who have access to multiple nodes and IP addresses. Alen _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
