EIP 908: Reward clients for a sustainable network
| Author | James Ray, Micah Zoltu |
|---|---|
| Discussions-To | https://ethereum-magicians.org/t/eip-908-reward-full-nodes-and-clients/241 |
| Status | Draft |
| Type | Standards Track |
| Category | Core |
| Created | 2018-03-01 |
A reward for running a full node is deprecated, but the proposal for a reward for clients remains
While Casper validators are incentivized to validate transactions, there are still no incentives for relaying blocks and storing data (which includes state). This paper is more a high-level analysis and discussion rather than attempting to provide a concrete solution. Pocket Network is a separate blockchain being designed as of Sept 2018 that incentivises relaying transactions, that is intended to be compatible with other blockchains. Note also that Rocket Pool is under development and is planned to be a pool for Casper, which will help to incentivise running a full node. Another alternative is VIPnode which charges fees to light clients for full nodes that serve them. In light of these solutions being developed, perhaps a more appropriate approach to generally rewarding clients would be to incentivize bandwidth (relaying and downloading), storage and I/O (while computation is already incentivized with gas for miners and will be for proposers under sharding and Casper). Note also that notaries will be incentivized to download collations under sharding. Outdated (Casper FFG will be implemented with Ethereum 2.0 with sharding: shasper): given that it looks like Casper FFG will be implemented soon, to minimize undue complexity to the protocol, incentivizing validation in the mean time may be considered not worthwhile. For a previous version of the proposal containing a proposal for rewarding a full node, refer to here.
Simple Summary
When each transaction is validated, give a reward to clients for developing the client.
Abstract
The tragedy of the commons is a phenomenon that is well known in many sectors, most notably in regard to sustainability. It involves the over-utilization of shared finite resources, which detriments all participants and stakeholders involved (which in the case of a global public good can be everyone, including future generations). Without proper management of public resources, a tragedy of the commons can occur. Internalizing externalities (where externalities can be broadly defined as effects that are not accounted for in the intrinsic price of a good, service or resource) is one way of incentivizing the proper management of resources, although other methods that actually properly manage them are necessary. This EIP proposes to make a change to the protocol to providea reward to clients for providing the software that enables Ethereum to function, where the reward can include a proportion of transaction fees (reducing the full proportion that the miner currently receives), and some newly minted ETH. Thus, clients are incentivized to maintain and improve the security and health of the Ethereum protocol and ecosystem. To summarize the mechanism in the proposal, a user agent is attached to a transaction, where this user agent contains a vector with the index of a client address in an access list. The client address could be inserted by the client and verified that it is the same as a read-only constant in the client’s storage.
Reward mechanisms that are external to being built in to the protocol are beyond the scope of this EIP. Such extra-protocol reward methods include state channel payments for extra services such as light client servers providing faster information such as receipts; state channel payments for buying state reads from full nodes; archival services (which is only applicable to future proposed versions of Ethereum with stateless clients); and tokens for the client and running full nodes.
Motivation
Currently there is a lack of incentives for anyone to run a full node, while joining a mining pool is not really economical if one has to purchase a mining rig (several GPUs) now, since there is unlikely to be a return on investment by the time that Ethereum transitions to hybrid Proof-of-Work/Proof-of-Stake with Casper FFG, then full PoS with CBC Casper.
Additionally, providing a reward for clients gives a revenue stream that is independent of state channels or other layer 2 mechanisms, which are less secure, although this insecurity can be offset by mechanisms such as insurance, bonded payments and time locks. Rationalising that investors may invest in a client because it is an enabler for the Ethereum ecosystem (and thus opening up investment opportunities) may not scale very well, and it seems that it is more sustainable to monetize the client as part of the service(s) that it provides.
While it may be argued that the funds raised by the Ethereum pre-mine (the pre-ICO and the ICO) can be used to fund client development, that argument is questionable, since Parity is VC-funded, and other clients such as Prysmatic Labs 1, 2, 3 and exthereum have received funding from other parties, whereas perhaps they would not have needed to do so if there was sufficient funding from the Ethereum pre-mine funds. Drops of Diamond is yet to receive any funding.
Incentivizing client development would more directly incentivize resource provision in the protocol, preventing a tragedy of the commons, where there is an extreme lack of supply and excess demand, leading to the protocol being unusable.
See here for an analysis in the context of Bitcoin, PoW, and a hybrid PoW/PoS protocol. While Ethereum has a gas limit, the section points out that this is not enough as the market cap increases and the incentive to attack the network increases, while the ratio of security costs to transaction fees does not, while PoS will further alleviate the problem. However, the section points out that PoS is not enough, since the costs of propagating, verifying and storing transactions are not incentivised. Note that the “Proof of Activity: Extending Bitcoin’s Proof of Work via Proof of Stake” paper also contains a scheme for incentivizing a target participation level.
The word “activity” in the phrase Proof of Activity emphasizes the point that only active stakeholders who maintain a full online node get rewarded, in exchange for the vital services that they provide for the network. This stands in contrast to earlier Proof of Stake schemes in which offline stake can accumulate weight over time, and may ultimately be utilized in double-spending attacks.
We can also incentivize full nodes to propagate transactions and to store transactions or state, in addtition to verifying them. While these first two incentivizations are outside the scope of this EIP, there are proposals here and here, respectively.
Implementing this as a layer 2 solution may not ensure the sustainability of the protocol, since not everyone would use it; if the protocol doesn’t have any cost for full nodes to validate transactions, then people will take advantage of that and not use the layer 2 solution. It seems that you should at least have the part where the reward is provided in protocol, but then that and the user agent signature doesn’t really add anything else to the protocol, so doing some part in-protocol and some part e.g. the verification or a verification-game off-protocol could be done, but it’s already done in protocol. Note also that some computationally expensive tasks are too challenging to feasibly do in protocol, e.g. due to not fitting in the gas limit, could be done with Truebit, where verifiers have an incentive.
Not providing incentives for clients is an issue now as there is less incentive to build a client that aligns with the needs of users, funds need to be raised externally to the protocol to fund client development, which is not as decentralized. If only a smaller subset is able to fund client development, such as VCs, angel investors and institutional investors, that may not align well with the interests of all current and potential stakeholders of Ethereum (which includes future stakeholders). Ostensibly, one of the goals of Ethereum is to decentralize everything, including wealth, or in other words, to improve wealth equality. Not providing incentives for full nodes validating transactions may not seem like as much of an issue now, but not doing so could hinder the growth of the protocol. Of course, incentives aren’t enough, it also needs to be technically decentralized so that it is ideally possible for a low-end mainstream computer or perhaps even a mobile or embedded IoT device to be a verifying full node, or at least to be able to help with securing the network if it is deemed impractical for them to be a full node.
Note that with a supply cap (as in EIP 960, the issuance can be prevented from increasing indefinitely. Alternatively, it could at least be reduced (still potentially but not necessarily to zero, or to the same rate at which Ether is burnt when slashing participants, such as validators under a Casper PoS scheme or notaries under a sharding scheme), e.g. by hard forks, or as per EIP 1015, an on-chain contract governed by a decision assembly that gets signalling from other contracts that represent some set of stakeholders.
Specification
Add a new field to each block called PrevBlockVerifications, which is an arbitrary, unlimited size byte array. When a client verifies that a previous block is valid, the client appends a user agent to PrevBlockVerifications via an opcode in a transaction, PREV_BLOCK_VERIF. The user agent is a vector with the immutable fields: the blockhash of the block that is validated, and the index of a client address in an access list (details are below). A miner validates a transaction before including it in a block, however they are not able to change these fields of the vector because they’re immutable.
Send 0.15 ETH to the client (see the rationale below), when the block is processed. The amounts could include a proportion of transaction fees (while the miner would then receive less), which would reduce newly issued ETH. These amounts are specified in new ClientReward fields in the block.
In order to only incentivize verifying recent blocks, assert that the block number corresponding to a blockhash is less than 400 blocks ago.
Attacks and added specifications to prevent them
A miner could create a client and fill their block with transactions that only contain the PREV_BLOCK_VERIF opcode (or alternatively, arbitrarily complex transactions, still with the opcode), with previous blockhashes that they have validated and the address of their client. To prevent this, state would have to store a list containing lists, with each sublist listing the blockhashes up to 400 blocks ago that correspond to a miner, then a miner (or a proposer under Casper CBC) could have to put down a deposit, and be slashed if the proposer inserts such a transaction (that contains a blockhash which they have already proposed, not just in a block but at any time previously). Updating the state to remove blockhashes more than 400 blocks ago would add additional overhead, although you could add an extra window for older blocks, say 120,000 blocks (roughly every 3 weeks), still ignore blocks that are older than 400 blocks ago, and remove these older 120,000 blocks every 120,000 blocks. An attacker could bribe a miner/proposer to include transactions like above that contain an address of a client in the access list which they own. However, the above slashing condition should disincentivize this.
More details on the access list
The access list prevents anyone inserting any address to the first element of the vector, where there may be a way to prevent censorship and centralization of authority of who decides to register new addresses in the list, e.g. on-chain governance with signalling (possibly similar to EIP 1015, which also specifies an alternative way of sending funds) or a layer 2 proof of authority network where new addresses can be added via a smart contract. Note that there may be serious drawbacks to implementing either of these listed examples. There is a refutation of on-chain governance as well as of plutocracy. Proof of Authority isn’t suitable for a public network since it doesn’t distribute trust well. However, using signalling in layer 2 contracts is more acceptable, but Vlad Zamfir argues that using that to influence outcomes in the protocol can disenfranchise miners from being necessary participants in the governance process. Thus, in light of these counterpoints, having an access list may not be suitable until a decentralized, trustless way of maintaining it is implemented and ideally accepted by the majority of a random sample that represents the population of Ethereum users.
However, another alternative to managing the access list would be to have decentralized verification that the address produced from querying an index in the access list does correspond to that of a “legitimate” client. Part of this verification would involve checking that there is a client that claims that this address is owned by them, that they are happy to receive funds in this manner and agree or arranged to putting the address in the access list, and that the client passes all tests in the Ethereum test suite. However, this last proviso would then preclude new clients being funded from the start of development, although such would-be clients would not be able to receive funds in-protocol until they implement the client anyway (as an aside, they could raise funds in various ways—a DAII, pronounced die-yee, is recommended, while a platform for DAIIs is under development by Dogezer). All of this could be done off-chain, and if anyone found that some address in the access list was not legitimate, then they could challenge that address with a proof of illegitimacy, and the participant that submitted the address to the access list could be slashed (while they must hold a deposit in order to register and keep an address in the access list).
Additionally, it should help with being only able to read the client’s address from the client, and the whole transaction could revert if the address is not in the access list. You could provide the index of the address in the access list, and then you could assert that the address found at that index matches that which can be read by the client (where the latter would be a read-only address).
Rationale
A rough qualitative analysis of fees
As of May 4 2018, there are 16428 nodes. Assume that an annual cost for an average client developer organisation is $1 million per annum. Projecting forward (and noting that the number of nodes should increase substantially if this EIP was implemented, thus aiding Ethereum’s goal of decentralizing everything) assume that there are 10 clients. Thus let us assume that the number of nodes doubles to 30000 nodes within 5 years (this assumption is probably conservative, even if it is forward looking). Assume for simplicity that the costs of a client are entirely covered by this block reward.
Average cost per client = number of nodes * Block reward per client / number of clients
Block reward per client (worse case for ETH price) = Average revenue per client * number of clients / (ETH exchange rate * number of nodes)
= $2,000,000 * 10 / (500 * 30,000)
= 1.333333333 ETH
Block reward per client (better case) = $1,000,000 * 10 / (2000 * 30,000) = 0.166666667 ETH
Suppose that we use a block reward of 0.15 ETH for clients.
More rationale (outdated by above)
The amount of computation to validate a transaction will be the same as a miner, since the transaction will need to be executed. Thus, if there would be transaction fees for validating full nodes and clients, and transactions need to be executed by validators just like miners have to, it makes sense to have them calculated in the same way as gas fees for miners. This would controversially increase the amount of transaction fees a lot, since there can be many validators for a transaction. In other words, it is controversial whether to provide the same amount of transaction fee for a full node validator as for a miner (which in one respect is fair, since the validator has to do the same amount of computation), or prevent transaction fees from rising much higher, and have a transaction fee for a full node as, say, the transaction fee for a miner, divided by the average number of full nodes validating a transaction. The latter option seems even more controversial (but is still better than the status quo), since while there would be more of an incentive to run a full node than there is now with no incentive, validators would be paid less for performing the same amount of computation.
And as for the absolute amounts, this will require data analysis, but clearly a full node should receive much less than a miner for processing a transaction in a block, since there are many transactions in a block, and there are many confirmations of a block. Data analysis could involve calculating the average number of full nodes verifying transactions in a block. Macroeconomic analysis could entail the economic security benefit that full nodes provide to the network.
Now, as to the ratio of rewards to the client vs the full node, as an initial guess I would suggest something like 99:1. Why such a big difference? Well, I would guess that clients spend roughly 99 times more time on developing and maintaining the client than a full node user spends running and maintaining a full node. During a week there might be several full-time people working on the client, but a full node might only spend half an hour (or less) initially setting it up, plus running it, plus electricity and internet costs. Full node operators probably don’t need to upgrade their computer (and buying a mining rig isn’t worth it with Casper PoS planning on being implemented soon).
However, on further analysis, clients would also get the benefit of a large volume of rewards from every full node running the client, so to incentivise full node operation further, the ratio could change to say, 4:1, or even 1:1, and of course could be adjusted with even further actual data analysis, rather than speculation.
Providing rewards to full node validators and to clients would increase the issuance. In order to maintain the issuance at current levels, this EIP could also reduce the mining reward (despite being reduced previously with the Byzantium release in October 2017 from 5 ETH to 3 ETH), but that would generate more controversy and discusssion.
Another potential point of controversy with rewarding clients and full nodes is that the work previously done by them has not been paid for until now (except of course by the Ethereum Foundation or Parity VCs funding the work), so existing clients may say that this EIP gives an advantage to new entrants. However, this doesn’t hold up well, because existing clients have the first mover advantage, with much development to create useful and well-used products.
There is a tradeoff. Higher fees means you may cut out poor people and people who just don’t want to pay fees. But if a laptop can run a full node and get paid for it then that would offset the fees through usage. Full nodes do provide a security benefit, so the total fees given could at least be some fraction of this benefit. Fees that go towards client development incentivise a higher quality client. To me, I think it makes more sense to internalize costs as much as possible: for computation, storage, bandwidth, I/O, client development, running full nodes, mining/validating, etc. You avoid a tragedy of the commons through externalizing costs. The more you internalize costs, the more sustainable it is, and the less you rely on rich people being generous, etc. (Although, getting philosophical, ultimately you can’t force rich people to be generous, they have to do so out of the kindness of their hearts.)
Regarding rewards for full nodes, in the draft phase 1 sharding spec proposers acting as full nodes have rewards for proposing blobs (without execution) or later in phase 3 transactions (with execution) to be included into collations/blocks. So that would help. However, full nodes that do not act as proposers and just verify transactions, or full state nodes, are still not incentivized.
Note that while further quantitative analysis to specify fees should be done, some level of experimentation after implementing this method on-chain may be necessary.
Security
All of the below struck out information should be prevented via using an access list and verifying that the read-only address provided by the client matches with an address in the access list, as well as using a layer 2 solution such as a PoA network for censhorship resistance and minimization of centralization in the access list.
Further discussion is at https://ethresear.ch/t/incentives-for-running-full-ethereum-nodes/1239.
Backwards Compatibility
Introducing in-protocol fees is a backwards-incompatible change, so would be introduced in a hard-fork.
Test Cases
TODO
Implementation
TODO
Copyright
Copyright and related rights waived via CC0.