I generally like the direction of this proposal in terms of allowing full nodes to run with a different storage/bandwidth tradeoff. Cory, were this implemented, would you expect Core to support both operating modes (full UTXO set and UHS) depending on user configuration, or would UHS be mandatory?

Also, given that Bram Cohen's TXO bitfield proposal was an inspiration for this, could you comment on why the UHS is preferable to that approach? An alternative that goes even further in the direction of more bandwidth, less storage, would be for nodes to simply maintain a Merkle Mountain Range over all TXOs in order of creation and a spentness bitfield. Blocks could be requested with the prev outputs and a Merkle proof linking them into the MMR root. Since the Merkle proof is deterministic, it could be computed by archive nodes and miners and saved alongside the block data for relay. Another benefit of this is the TXO MMR root may be independently useful if committed into the coinbase transaction.

On Thu, Jun 7, 2018 at 7:02 AM Sjors Provoost via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
eMMC storage, which low end devices often use, come in 2x increments. Running a pruned full node on 8 GB is difficult if not impossible (the UTXO set peaked at 3.5 GB in January, but a full node stores additional stuff).

However, 16 GB is only €10 more expensive and presumably standard by the time this would be rolled out.

On AWS every GB of SSD storage avoided saves $1 per year, not end of the world stuff, but not negligible either. Outbound traffic costs $0.10 / GB (ignoring free allowance), so when uploading 200 GB per year, the 5% would offset $1 of storage cost savings.

The above seems marginal, probably not worth it unless there’s really no downside.

What I find attractive about this proposal is the ability to squeeze more out of limited RAM (typically only 1 or 2 GB on these low end devices). I’d have to test Cory’s branch to see if that actually matters in practice.

It’s also useful to distinguish benefits during initial sync from ongoing operation. The former I’ve almost given up on for  low end devices (can take weeks), in favor of doing it on a faster computer and copying the result. The latter needs far less RAM, so perhaps this proposal doesn’t help much there, but that would be useful to measure.

Did you try the recent SHA256 optimizations on your branch?

Sjors

> Op 17 mei 2018, om 18:56 heeft Gregory Maxwell via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> het volgende geschreven:
>
> On Wed, May 16, 2018 at 4:36 PM, Cory Fields via bitcoin-dev
> <bitcoin-dev@lists.linuxfoundation.org> wrote:
>> Tl;dr: Rather than storing all unspent outputs, store their hashes.
>
> My initial thoughts are it's not _completely_ obvious to me that a 5%
> ongoing bandwidth increase is actually a win to get something like a
> 40% reduction in the size of a pruned node (and less than a 1%
> reduction in an archive node) primarily because I've not seen size of
> a pruned node cited as a usage limiting factor basically anywhere. I
> would assume it is a win but wouldn't be shocked to see a careful
> analysis that concluded it wasn't.
>
> But perhaps more interestingly, I think the overhead is not really 5%,
> but it's 5% measured in the context of the phenomenally inefficient tx
> mechanisms ( https://bitcointalk.org/index.php?topic=1377345.0 ).
> Napkin math on the size of a txn alone tells me it's more like a 25%
> increase if you just consider size of tx vs size of
> tx+scriptpubkeys,amounts.  If I'm not missing something there, I think
> that would get in into a very clear not-win range.
>
> On the positive side is that it doesn't change the blockchain
> datastructure, so it's something implementations could do without
> marrying the network to it forever.
>

_______________________________________________
bitcoin-dev mailing list
bitcoin-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev