On 27/04/2014 02:05 p.m., Mark Friedenbach wrote:


On 04/27/2014 05:36 AM, Sergio Lerner wrote:

Without invoking moon math or assumptions of honest peers and
jamming-free networks, the only way to know a chain is valid is to 
witness the each and every block. SPV nodes on the other hand,
simply trust that the most-work chain is a valid chain, based on
economic arguments about the opportunity cost of mining invalid
blocks.

I argue that you cannot talk about "the most-work chain" without 
actually making an assumption about honest peers.

I should have said "without invoking moon math or interactive protocols
requiring honest peers over jamming-free networks." The interactive
protocol was more the point than the honest peers and jamming-free
network. Yes, without an honest peer and an un-jammed network, you might
never learn about the most-work chain in the first place. But having the
security of the proof not depend on query access to an honest full node
is absolutely necessary for some applications and certainly desirable in
others.

The problem is not having or not access to a honest full node. The SPV client MUST have access to a honest full node sometime.
The problem is WHEN. One can make the security assumption that during an attack (someone gives you a fake block) you also loose the possibility to reach any honest node. Then SPV proofs come into play.

Here are the security assumptions I added to my post about SmartSPV to clarify this:

Security Assumptions

First let’s review the main security assumption of headers-only SPV:

The attacker does not control all your communications with the payment network.

This means that there is at least a single connected peer that behaves honestly. This assumption is quite strong and may not hold during short periods of time, such as during application power-on (when only a few peers have been connected), or when moving to a place where the ISP is untrusted. For SmartSPV we’ll require weaker security assumptions:

The attacker cannot control all your communications with the payment network for more than a fixed period of time (e.g. 2016 blocks for Bitcoin or approximately 15 days)
The attacker is rational: it won’t spend an huge amount of money to steal a much smaller amount.

This assumptions imply that the attacker may control all your Internet connections while he sends you a malicious block branch containing a fake payment to you.

First this is a method that uses NPPs, not SPV proofs. Since the
method chooses all peers that are synchronized (have the latest
current block) then going back means only skipping a potential short
fork (which I suppose has never been more than 3 blocks during normal
network conditions). You're looking for a common ancestor, not the
checkpoint. So the linear scan is actually O(1). The exact value can
be approximated by the sum of the convergent infinite geometrical
sequence of forking probabilities, which it's about 1.03 without
considering selfish-mining, and probably less than 2.03 considering
it.

Unless you're connected to attacker nodes which are wildly divergent
from each other. It's relatively easy to create a massive fake history
of difficulty-1 blocks.

Since in my use case (SmartSPV) I proposed you start from the most recent block and go backwards, the attacker must compete in PoW with the real current difficulty informed.
Suppose the SPV client looks for 6-block chains backwards starting from the last current block. Suppose you know or estimate the current network difficulty. Suppose a malicious peer creates a fake 6-block chain Cm and the honest peer gives you the 6-block chain Ch. If Ch has not the expected work it's discarded. If both has the expected work, then you better not true any of them and walk their parents until you find a common parent. That's the block you should trust. If you don't have an honest node connected, then the only decide to trust or not Cm is by it's accumulated work (and you have already a bound for it)

If you assume honest peers things get very easy. But that's not a safe
assumption to be making. With back-link block-history commitments, on
the other hand, an interactive protocol allows you to do a binary search
to find common ancestors, and have trust that the intermediate links
actually exist.

So you agree that: you need a periodic connection to a honest node, but during an attack you may loose that connection. This is the assumption we should be working on, and my use case (described in http://bitslog.wordpress.com/2014/04/25/smartspv-a-better-simplified-payment-verification-for-smartphones/) assumes that.