[Bitcoin-development] Blockchain as root CA for payment protocol

[Timo Hanke]

On Sat, Feb 09, 2013 at 07:01:48PM +0000, Luke-Jr wrote:
> On Saturday, February 09, 2013 2:33:25 PM Timo Hanke wrote:
> > namcoin tries to solve a different problem, DNS, whereas I want
> > to establish an identity for a payment protocol.
> 
> What is the technical difference here? Namecoin ties names to data; DNS is a 
> specific namespace in it. There is no reason I know of that this identity 
> stuff cannot be a new namespace.

It's not about technical differences, but about the different use or
purpose, which can result in different security demands. I argue that
DNS has a lower demand in this respect than payment ids have. So DNS
data can be in a chain with a hashrate lower than bitcoin's hashrate but
payment ids _for_ bitcoin have to be in a chain with equal hashrate.

> > You can argue that alt-chains _can_ be as strong as bitcoin, but they
> > don't _have to_ be. There is no guarantee how many people will
> > cross-mine.
> 
> This is true of namecoin, but it does not have to be true of new merged-mined 
> data. You could very well require the Bitcoin proof-of-work to be valid and 
> the master header to be in the Bitcoin blockchain.

Ok, true. This does the trick. If few miners merge-mine then the new
chain just becomes slower. But is this still an alt-chain? It is not
independently verifiable anymore, like the alt-chains described in the
wiki are. Instead, you need to refer to the bitcoin's chain to see if
the target is correct. Not sure if I got you right on this. But it seems
to be essentially a more efficient version of what I proposed, rather
than a true alt-chain.

I suppose you suggest to place the master header hash into the coinbase.
A drawback may be that it puts miners at a great advantage over regular
users. This could (but doesn't have to) become relevant depending on
your counter-measures against excessive alias registration. I think
Peter addressed this (below). 

On Fri, Feb 08, 2013 at 06:01:08AM -0500, Peter Todd wrote:
> Finally, why is this implemented within the reference client? Use the
> raw transaction API and make up your own database. If you want, create a
> RPC command that allows you to query the UTXO set directly; this would
> be a useful feature to have. This patch will never be accepted to the
> reference client, so you'll just wind up having to maintain a fork. Even
> for a prototype this approach is ill-advised - prototypes have a bad way
> of turning into production code.

This was not intended to be a prototype and will certainly not be
maintained. It is a demo to be run on the testnet to get a feel of how
the user interface (RPC) and the "work flow" could look like, starting
from the creation of a certificate all the way to paying to a
customer-derived payment addresses (pay-to-contract) when the merchant's
base address is defined in the certificate. There's an appeal to be able
to issue
./bitcoind sendtoalias foo deadbeef 10
and being sure that 10 BTC go, e.g., to a unique P2SH multisig address
that is derived for order number 'deadbeef' from the two pubkeys that
foo defined in his certificate. And having the certificate verification
happen automatically in the background. The demo is in the reference
client a) to simulate this feel, b) because it was the fastest way to
code it. Apart from that, it could have just as well been separate, and
an UTXO query-RPC would certainly be nice.  

Another reason for this demo was the fun of devising a certificate that
can handle all this. 

BTW, I'm sure that some form of certificate handling will find its way
directly into the reference client. The user will want to trust only one
piece of software running on one piece of dedicated hardware. 

On Mon, Feb 11, 2013 at 06:21:03AM -0500, Peter Todd wrote:
> In what way are you not solving the same problem as DNS? I don't mean
> the Luke-Jr's (quite correct) technical point about key-value maps, I
> mean the human problem that I have these unique numbers that I can't
> memorize, and I have some non-unique names that I can.
> 
> By creating Yet Another Totally Different System you are just creating
> another way that users can be confused into thinking some name snatched
> up by some scammers in some little-used PKI system is who they are
> supposed to be communicating with. Fortunately your PKI system isn't
> actually used and probably never will be, so it's not a big deal yet,
> but ultimately you are adding to the problem.

What exactly is the problem, the "little-used"? Otherwise it's the same
as it is now, you pick up an interesting domain name on the street, type
it in, and start communicating with who you think it is, and maybe even
pay them. The EV that centralized PKI offers prevents only some attacks,
and may even create a false sense of security.

> Go work on namecoin and make it more usable. Then add some PKI to it
> using the *same* domain names so when I see a PKI certificate for "foo"
> I know it must be the same "foo" website I just visited and the same
> "foo at foo" I just emailed.

No. I outlined above why I think namecoin is unsuitable (but Luke's
suggestion for a new alt-chain may be). BTW, if you want to link aliases
of some "bitcoin PKI" and "domain names", they should not correspond
bijectively to each other. The "bitcoin alias" corresponds to a payment
base address, which I see as a more universal identifier than a domain
name. First, bitcoin alias foo can have several domains foo.com,
foo.co.uk, etc. but still only one "bitcoin certificate". This can be
achieved by either writing the domain names directly into the
certificate, or better, by having the "bitcoin certificate" sign an SSL
sub-certificate and tell your browser to ask bitcoind to verify it.
Second, a bitcoin alias can be meaningful without any domain names, e.g.
it can be a certificate for all vending machines of a certain seller,
with who you interact directly, say NFC. To summarize, I would either
not link bitcoin aliases with domain names, or put SSL certificates
"below" bitcoin certificates in the verification chain.

Also BTW, it is the beauty of the pay-to-contract principle that
authenticated communication is not even required. This means that if you
know you have the correct "bitcoin certificate" stored with your
bitcoind and you pay to it, then it doesn't matter if you ordered on a
wrong or fake website. The worst-case scenario is that you have to call
in via another channel and re-submit your order, but funds are never
lost.

> Alt-chains don't have to be based on mining you know. Your proof-of-work
> can be replaced by proof-of-sacrifice, specifically Bitcoins. 
> [...]
> 1) https://github.com/petertodd/trustbits/blob/master/fidelitybond.md

Yes, I already read your discussing of sacrifice-methods after your
first reply and found it interesting.. The problem I see is to
dynamically (and automatically) adjust the minimum sacrifice amount. A
strict limit on the number of registrations like namecoin has is not
desirable. A constant sacrifice, as you mentioned earlier, is also
undesirable. A good measure could be the average transaction fee taken
over several blocks. Maybe the minimum sacrifice should be a constant
times that. 

> > You are probably right that storing this in the _spent outputs_
> > would be better [than UTXO]. There doesn't seem to be any type of
> > client out there that would benefit from having to search UTXO only. 
> 
> The blockchain grows at a maximum rate of 55GiB/year. Do you think your
> users will all want to have that available just to validate some PKI
> certificates?

Actually yes. They are also willing to do that to verify their payments,
so why not for the certificate? But true, this was the reason why I
thought of putting it in the UTXO. To enable some future dedicated
"hardware wallet" to keep only the UTXO and to verify against it. Where
would you store you alt-chain block header hashes? UTXO? 

BTW, suppose you avoid any squatting (there would certainly be ways to
do that). Then with my original proposal of one transaction per
registration you would grow the UTXO by O(n), n is the number of users.
Each user would create a small constant number of aliases. This number
will certainly be dwarfed by the number of unspend outputs that each
user keeps anyway as a result of their regular transactions. So it would
not present a problem for the UTXO. Actually, anything above the dust
threshold should not be a problem, or the system is misdesigned.  But of
course the alt-chain header hashes would only take O(1), much better.

Timo