[Bitcoin-development] Cartographer

[Mike Hearn]

>
> Can you explain further where limitations and problems were hit?
>

Well, look at the feature list :)

The biggest need from my POV is querying support. It's awkward to try and
retrofit flexible key=value pair queries onto DNS, it just wasn't designed
for that. With HTTP it's easy. This will become more important in future as
the protocol evolves. For example, some nodes will soon stop serving the
block chain because they start pruning. Today this is managed with a hack:
pruning nodes just stop providing *all* services to the p2p network. This
takes them out of the DNS seeds entirely. But they can actually still
provide download of the parts of the chain they still have, and they can
provide transaction filtering and relay support, along with misc other
things.

With the current DNS protocol you get an all or nothing choice - so
probably seeds that only support it will elect to only show nodes that have
the entire block chain, because that's what Bitcoin Core would find most
useful. SPV wallets have slightly different needs.

In theory you could come up with a pile of hacks to specify what the client
needs in the DNS query, but then you have a v2 protocol anyway and might as
well go the whole way.

Additionally, with DNS it's awkward to provide extra data in the responses
beyond IP address and it's VERY awkward to sign the responses. Signing the
responses has a couple of benefits. The biggest is, in future it'd be nice
to have an authenticated and encrypted network, to raise the bar for sybil
and MITM attacks. DNS seeding can't be upgraded to support that with any
reasonable level of effort. And DNS is awkward to configure/set up.
Actually DNS is just awkward, period.

The second benefit of signing is it provides audibility. If you see a seed
give bad answers, you can now prove it to other people.

There is also the previously discussed issue that DNS seeds sometimes get
blocked by aggressive networks because they start serving IPs that are
infected with malware i.e. they look like fast-flux sites.

Using a simple HTTP based protocol fixes all of these problems in one go.

Now, re: privacy.

Firstly, Peter, get help. I'm serious. You are starting to sound like an
auto-generated parody of yourself. When you can't debate something as
boring as HTTP vs DNS without trying to raise an angry mob using bizarre
conspiracy theories, that's not normal.

I don't think the "DNS has caches" issue is worth worrying about for a lot
of reasons:

   1. Full nodes try as hard as they can to open ports and advertise their
   IP addresses to everyone. Even if you change the defaults to disable that,
   you're about to connect to a bunch of random computers with no reputation
   anyway.

   2. Lists of stale IP addresses are hardly useful to regular people and
   network operators can identify Bitcoin users by looking for traffic on port
   8333, so it's unclear what threat model is being addressed here.

   3. The biggest users of the seeds are all SPV wallets. Every single one
   of these already phones home to check for online updates.

   4. DNS proxying only hides part of the IP address. If you're serious
   about this you want to be doing lookups via Tor. Whilst it's possible to
   use the DNS seeds via Tor in a reasonable way with exit diversity (and
   bitcoinj does), doing it requires low level Tor protocol programming
   <https://github.com/bitcoinj/bitcoinj/blob/42f9d7c193fcd56fda7691b0ea934bae9d23f2d6/core/src/main/java/org/bitcoinj/net/discovery/TorDiscovery.java#L260>
   that is out of reach for most implementors. An HTTP lookup is trivial with
   any HTTP stack that supports SOCKS.

   5. ISPs also deploy transparent HTTP caches. The Cartographer protocol
   uses HTTP with inline signing so responses can be cached, once the right
   headers are being set.

tl;dr it's unclear that DNS caching actually solves any real privacy
problem but regardless, you can get the same distributed caching with HTTP
as with DNS. So in the end it makes little difference.

I believe that Cartographer is a better protocol all round and there are no
costs beyond the one-time upgrades, but even if for some reason you
disagree with the five privacy points above, I think the benefits still
massively outweigh the costs.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfoundation.org/pipermail/bitcoin-dev/attachments/20141229/c1277067/attachment.html>