The whole point of having headers built at a constant size and generation rate is to minimize the amount of data needed to "understand" of the blockchain while simultaneously maximizing integrity/security in the presence of untrusted nodes.  Barring the 50%-attack, you only need a couple honest nodes out of 50 to stay safe (as long as you're waiting for your 6 confirmations).   In fact, I would argue that a full node (Satoshi client), has the same level of security as a headers-only client... because they both base all their verification decisions on computations that end with comparing hashes to the longest-chain headers.

In the case that an attacker figures out how to isolate your node entirely and start feeing you poisoned blocks, then you are vulnerable with any kind of node, full or lightweight.  I don't see where the reduced security is. 

The only issue I see is that a truly light-weight, headers-only node will be having to download an entire block to get one transaction it needs.  This would be significantly alleviated if nodes can start requesting merkle-trees directly, even without merkle-branch-pruning.   If a node can ask for a tx and the tx-hash-list of the block that incorporated that tx,  he can easily verify his tx against his no-need-to-trust-anyone headers, and doesn't have to download MBs for every one. 

As for blockchain pruning... I think it's absolutely critical to find a way to do this, for all nodes.  I am swayed by Dan Kaminsky's scalability warnings, and my instinct tells me that leaving full verification to a select few deep-pockets nodes in the future opens up all sorts of centralization/power-corporation issues that is contrary to the Bitcoin concept.  It is in everyone's best interest to make it as easy as possible for anyone to act as a full node (if possible).  As such, I believe that the current system of minimizing TxOut size is the right one.  TxIns take up 0 bytes space in the long-run when taking into account any blockchain pruning/snapshot idea (except for nLocktime/seq transactions where the TxIn might have to be saved). 

-Alan





On 12/18/2011 12:09 PM, theymos wrote: 
On Sat, Dec 17, 2011, at 05:27 PM, Jordan Mack wrote:

I don't like the idea of a header only client, unless this is just an
interim action until the full block chain is downloaded in the
background. Development of these types of clients is probably
inevitable, but I believe that this breaks the most fundamental
aspects of Bitcoin's security model. If a client has only headers, it
cannot do full verification, and it is trusting the data from random
anonymous peers.

A headers-only client is much better than trusting anyone, since an
attacker needs >50% of the network's computational power to trick
such clients.

For everyone to keep being a full node, hardware costs would need to
constantly go down enough for all nodes to be able to handle enough
transactions to meet demand. If hardware doesn't become cheap enough
quickly enough, either some people would be unable to handle being full
nodes, or the max block size wouldn't rise enough to meet demand and
transaction fees would become noncompetitive.

------------------------------------------------------------------------------
Learn Windows Azure Live!  Tuesday, Dec 13, 2011
Microsoft is holding a special Learn Windows Azure training event for 
developers. It will provide a great way to learn Windows Azure and what it 
provides. You can attend the event by watching it streamed LIVE online.  
Learn more at http://p.sf.net/sfu/ms-windowsazure
_______________________________________________
Bitcoin-development mailing list
Bitcoin-development@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bitcoin-development