Peter Todd via bitcoin-dev
2017-02-23 01:11:47 UTC
Something I've recently realised is that TXO commitments do not need to be
implemented as a consensus protocol change to be useful. All the benefits they
provide to full nodes with regard to allowing for old UTXO data to be pruned -
and thus solving the UTXO bloat problem - can be implemented even without
having miners commit to the TXO commitment itself. This has a significant
deployment advantage too: we can try out multiple TXO commitment schemes, in
production, without the need for consensus changes.
1) Like any other merkelized data structure, a TXO commitment allows a data set
- the TXO set - to be securely provided by an untrusted third party, allowing
the data itself to be discarded. So if you have a valid TXO commitment, you can
discard the TXO data itself, and rely on untrusted entities to provide you that
data on demand.
2) The TXO set is a super-set of the UTXO set; all data in the UTXO set is also
present in the TXO set. Thus a TXO commitment with spent TXO's pruned is
equivalent to a UTXO set, doubly so if inner nodes in the commitment tree
commit to the sum-unspent of their children.
3) Where a outpoint-indexed UTXO set has a uniform access pattern, an
insertion-ordered TXO set has a delibrately *non-uniform* access pattern: not
only are new entries to the TXO set always appended to the end - an operation
that requires a known, log2(n), sized set of merkle tips - but due to lost
coins alone we can guarantee that older entries in the TXO set will be less
frequently updated than newer entries.
4) Thus a full node that doesn't have enough local storage to maintain the full
UTXO set can instead keep track of a TXO commitment, and prune older UTXO's
from it that are unlikely to be spent. In the event those UTXO's are spent,
transactions and blocks spending them can trustlessly provide the necessary
data to temporarily fill-in the node's local TXO set database, allowing the
next commitment to be calculated.
5) By *not* committing the TXO commitment in the block itself, we obsolete my
concept of delayed TXO commitments: you don't need to have calculated the TXO
commitment digest to validate a block anyway!
# Deployment Plan
1) Implement a TXO commitment scheme with the ability to efficiently store the
last n versions of the commitment state for the purpose of reorgs (a
reference-counted scheme naturally does this).
2) Add P2P support for advertising to peers what parts of the TXO set you've
3) Add P2P support to produce, consume, and update TXO unspentness proofs as
part of transaction and block relaying.
# Bootstrapping New Nodes
With a TXO commitment scheme implemented, it's also possible to produce
serialized UTXO snapshots for bootstrapping new nodes. Equally, it's obviously
possible to distribute those snapshots, and have people you trust attest to the
validity of those snapshots.
I argue that a snapshot with an attestation from known individuals that you
trust is a *better* security model than having miners attest to validity: the
latter is trusting an unknown set of unaccountable, anonymous, miners.
This security model is not unlike the recently implemented -assumevalid
scheme(1), in that auditing the validity of the assumed valid TXO commitments
is something anyone can do provided they have a full node. Similarly, we could
ship Bitcoin nodes with an assumed-valid TXO commitment, and have those nodes
fill in the UTXO data from their peers.
However it is a weaker security model, in that a false TXO commitment can more
easily be used to trick a node into accepting invalid transactions/blocks;
assumed valid blocks requires proof-of-work to pull off this attack. A
compromise may be to use assumed valid TXO commitments, extending my partial
UTXO set(2) suggestion of having nodes validate the chain backwards, to
eventually validate 100% of the chain.
2) [Bitcoin-development] SPV bitcoind? (was: Introducing BitcoinKit.framework),
Peter Todd, Jul 17th 2013, Bitcoin development mailing list,