To: bitcoin-dev@lists.linuxfoundation.org
From: Chris Belcher <belcher@riseup.net>
Message-ID: <bf9d6c9d-7cd4-bdce-c4f9-aacc09d4a06e@riseup.net>
Date: Wed, 30 Aug 2017 17:15:56 +0100
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Subject: [bitcoin-dev] Payment Channel Payouts: An Idea for Improving P2Pool
	Scalability
Precedence: list

Pooled mining in bitcoin contributes to miner centralization. P2Pool is
one solution but has bad scalability; additional hashers require the
coinbase transaction to be larger, bigger miners joining increase the
variance of payouts for everyone else, and smaller miners must pay extra
to consolidate dust payouts. In this email I propose an improved scheme
using payment channels which would allow far more individual hashers to
mine on p2pool and result in a much lower payout variance.


== Intro ==

P2Pool is a decentralized pool that works by creating a P2P network of
hashers. These hashers work on a chain of shares similar to Bitcoin's
blockchain. Each hasher works on a block that includes payouts to the
previous shares' owners and the node itself. The point of pooling is to
reduce the variance of payout, even though on average the reward is the
same (or less with fees). The demand for insurance, and the liquid
markets for options show that variance does have costs that people are
willing to pay to avoid.

Here is an example of a p2pool coinbase transaction:
https://blockchain.info/tx/d1a1e125ed332483b6e8e2f128581efc397582fe4c950dc48fadbc0ea4008022

It is 5803 bytes in size, which at a fee rate of 350 sat/b is worth
0.02031050 btc of block space that p2pool cannot sell to any other
transaction. As bitcoin inflation goes down and miners are funded more
by fees, this puts p2pool at more and more of a disadvantage compared to
trusted-third-party mining pools.

As each hasher is paid to their own bitcoin address, this limits the
number of hashers taking part as adding more individual people to the
payout transaction increases its size. Also small payouts cost a
disproportionate amount in miner fees to actually spend, which hurts
small miners who are essential to a decentralized mining ecosystem.

This could maybe be solved by keeping a separate balance state for each
user that is independent from the payouts, and make payouts only when
that balance state exceeds some reasonable threshold. But this increases
the variance which goes against the aim of pooled mining.


== Payment Channels ==

What's needed is a way to use off-chain payments where any number of
payments can be sent to each individual hasher without using the
blockchain. Then the N of the pay-per-last-N-shares (PPLNS) of p2pool
can be increased to something like 6-12 months of shares and so as long
as a small miner can mine a share every few months they will always get
a payout when p2pool finds a block. The payment channels would be in a
hub-and-spokes system and would work in a similar way to coinswap,
lightning network, atomic cross-chain swaps or any other contract
involving hashlocks and timelocks.

There would still be a sharechain but with hashers paying the entire
block reward to a hub. This hub would have a one-way payment channel
open to every hasher in p2pool and there would be created a situation
where if the hub gets paid then the hashers cannot fail to get paid.
Because cheating is impossible, the hub and hashers will agree to just
release the money to each other without resorting to the blockchain.

The coinbase address scriptPubKey would be this, block rewards are paid
to here:

     2of2 multisig
     hub + successful hasher
  OR
     hub pubkey + H(X)
  OR
     successful hasher pubkey + OP_CSV 6 months

A 2of2 multisig between the hub and the "successful" hasher which found
the block, although with a hashlock and timelock. H(X) is a hash value,
where the preimage X is generated randomly by the hub and kept secret,
but X will be revealed if the hub spends via that execution path. The
OP_CSV execution path is there to stop any holdups or ransom, in the
worst case if the hub stalls then the successful hasher can steal the
entire coinbase as punishment after 6 months.

Each payment channel address has this scriptPubKey:

     2of2 multisig
     hub-C + hasher-C
  OR
     2of2 multisig + H(X)
     hub-U + hasher-U

The pubkeys hub-C/hasher-C refer to 'cooperative' pubkeys.
Hub-U/hasher-U refer to 'uncooperative' pubkeys. Before a hasher starts
mining the hub will open a one-way payment channel to the hasher, and
pays some bitcoin to it (let's say 0.5btc for example).

The hashers mine a sharechain, a solved share contains the hasher's
cooperative and uncooperative pubkey. The hub keeps up with the
sharechain and announces partially-signed transactions going to each
hasher. The transactions are updated states of the payment channel, they
pay money to each hasher in proportion to the work that the hasher
contributed to the sharechain. The transaction contains a signature
matching the hub-U pubkey, the hasher could sign it with their hasher-U
key and broadcast except they still need the value of X.

If a hasher is successful and finds a share that is also a valid bitcoin
block, they broadcast it to the network.

Now, the hub can spend the block reward money on its own but only by
revealing X. Each hasher could then take that X and combine it with the
partially-signed transaction and broadcast that to get their money. So
if the hub gets paid then the hashers cannot fail to get paid. Since
defecting is pointless, the hub signs the hub-C signature of the
partially-signed transaction and sends it to each hasher, then the
successful hasher signs the 2of2 multisig sending the block reward money
to the hub. The successful hasher gets a small bonus via an updated
payment channel state for finding the block, to discourage withholding
same as today's p2pool.

These payment channels can be kept open indefinitely, as new blocks are
found by p2pool the hub creates new partially-signed transactions with
more money going to each hasher. When the hasher wants to stop mining
and get the money, they can add their own hasher-C signature and
broadcast it to the network.

If there's ever a problem and the hub has to reveal X, then all the
payment channels to hashers will have to be closed and reopened with a
new X, because their security depends on X being unknown.


== Hubs ==

The hub is a central point of failure. It cannot steal the money, but if
it gets DDOS'd or just becomes evil then the whole thing would stop
working. This problem could be mitigated by having a federated system,
where there are several hubs to choose from and hashers have payment
channels open with each of them. It's worth noting that if someone has a
strong botnet they could probably DDOS individual p2pool hashers in the
same way they DDOS hubs or even centralized mining pools.

The hub would need to own many bitcoins in order to have payment
channels while waiting for blocks to be mined. Maybe 50 times the block
reward which today would be about 650 bitcoins. The hub should receive a
small percentage of each block reward to provide them with an incentive,
we know from JoinMarket that this percentage will probably be around
0.1% or less for large amounts of bitcoin. Prospostive hub operators
should write their bids on a forum somewhere and have their details
added to some list on github. Hashers should have an interface for
blacklisting, whitelisting, lowering and raising priority for certain
hubs in case the hub operators behave badly.

As well as the smart contract, there are iterated prisoner's dilemma
effects between the hub and the hashers. If the hub cooperates it can
expect to make a predictable low-risk income from its held bitcoins for
a long time to come, if it does something bad then the hashers can
easily call off the deal. The hub operator will require a lot of profit
in order to burn its reputation and future income stream, and by
damaging the bitcoin ecosystem it will have indirectly damaged its own
held bitcoins. A fair pricing plan will probably have the hub taking a
small percent to start with and then 12 months later that percentage
goes up to take into account the hub's improved reputation.


== Transaction Selection ==

All the hashers and hub need to know the exact value of the block reward
in advance, this means they must know what the miner fees will be. This
is probably the most serious problem with this proposal.

One possible way to solve this is to mine transactions into shares and
so use the sharechain to make all the hashers and hubs come to consensus
about exactly which transactions they will mine, and so exactly what the
total miner fee will be. A problem here is that this consensus mechanism
is slow, immediately after a bitcoin block is found all the p2pool
hashers will have to wait 30-120 seconds before they know what
transactions to mine, so this would make them uncompetitive as a mining
operation.

Another way to deal with this is to have the hub just choose all the
transactions, announcing the transactions, total miner fee and merkle
root for the hashers to mine. This would work but allows the hub to
control and censor bitcoin transactions, which mostly defeats the point
of p2pool as an improvement to bitcoin miner centralization.

Another way is to have the hashers and hub estimate what the total miner
fee value will be. The estimate could start from the median miner fee of
the last few blocks, or from the next 1MB of the mempool. The hub would
announce all the partially-signed transactions to every hasher, and then
periodically (say every 60 seconds) announce updated versions depending
on how the mempool changes. Let's analyze what happens if the estimated
and actual rewards are different. If the actual block reward is lower
than the estimated reward, then the hub can update the payment channel
state to slightly lower values to take that into account when it
announces the cooperative hub-C signatures. The hashers can't use the
higher channel state without knowing X. The successful hasher will get
their bonus for finding the block which should help in encouraging them
to actually sign the hub's payout transaction. If the actual block
reward is higher than the estimated reward, the hub would hopefully
still update the hasher's payment channel states because of the
interated-prisoners-dilemma effects. But if the actual reward is much
higher then the hub may find it profitable to burn its reputation and
take the money by revealing X, one situation where this might happen is
if someone accidently pays a very high miner fee and a hasher mines it
without it being taken into account in the hub's regular payment channel
state updates. Apart from that very specific situation, this scheme of
estimating the total miner fee should work.


== Some Notes ==

*) Block rewards are locked for 100 blocks before they can be spent, so
the cooperative signatures should be exchanged after 100 blocks just in
case the block gets made stale/orphaned. While the hashers are waiting
for the 100 reward maturity period, they should mine with another hub as
the payout.

*) Today's p2pool has a feature for donating to individual hashers, this
could be replicated in the payment channel system by having each share
also contain the hasher's bitcoin address for donations (or possibly
their LN payment code)

*) Each hasher should probably be made to pay some bitcoins into the
payment channel address too, to stop DOSers locking up all the hub's
bitcoins. If the hasher doesn't find a share within some timeout then
the hub should close the payment channel.

*) Now that we have segwit all these payment channel schemes are much
easier to code.

*) The hashers must keep their money locked up in the payment channel
for months before enough collects. This could be a problem because some
miners don't really want to hold bitcoin long term. I wonder if theres
some way to link up these channels to LN so they can sold straight away.
They could also use futures contracts to sell the coins today at a
discount and actually deliver the coins later when they close the channel.


== References ==

*) https://en.bitcoin.it/wiki/P2Pool how p2pool works

*) https://bitcointalk.org/index.php?topic=18313.msg13057899#msg13057899
the scaleability problems of p2pool

*) https://bitcointalk.org/index.php?topic=18313.msg20943086#msg20943086
making the PPLNS window longer

*) book: The Evolution of Co-Operation by Robert Axelrod, for explaining
iterated prisoner's dilemma effects in detail

Thanks to the p2pool developer veqtrus for reviewing this