State growth and GasToken smart contract

Analysis of the state growth issue caused by GasToken smart contract, in Ethereum Classic network.

Recently, a contract called GasToken has spiked in usage on Ethereum Classic network, causing the state size to grow several GBs per day. Community members have been trying to propose solutions to fix the issue or at least to slow down the growth. This article provides some reviews and feedbacks on those proposed solutions.

Reduce block gas limit

The major issue of contracts like GasToken is that it causes fast state size growth. One of the simplest method to limit state size growth is to simply reduce the block gas limit. With less state access opcodes able to execute at each block, less data can be added into the state, and thus slower state growth.

However, the counterintuitive but unpleasant fact is, when the limit is decreased, miners may be more incentivzed to mine GasToken. The reason for this is that as the block size decreases, people will expect an increase in gas price shortly after. This is exactly the situation GasToken takes advantage of. As a result, user experiences in the beginning will be worse, because we can expect a larger portion of the block to be filled with GasToken transactions, and users will have to pay higher gas price to get their non-GasToken transactions through.

Remove gas refund

GasToken takes advantage of the EVM mechanic of gas refund. As a result, some community members started to talk about removing the gas refund mechanic so as to break GasToken contract, halting the state growth. Reviewing this method, we find it genuinely breaking more things than accomplishing the goal.

Gas refund does not fundamentally contribute to the issue of state growth. The actual issue of state growth is due to opcodes' utilities, rated at their absolute costs. While gas refund mechanic is a part of the calculation for absolute costs, it is not the source. In addition, we note that gas refund provides incentive for clearing state. If we remove gas refund without other accomodating changes, it will also leave abnormality in the absolute costs. What’s more, GasToken contracts' utility of "save gas now, spend it later" can be fulfilled by specialized smart contracts without using the gas refund mechanic.

Mechanic

Gas refund is a way to remove gas cost for an operation. It presents on two opcodes:

SSTORE: Refund 15000 gas, if a storage value is reset.
SELFDESTRUCT: Refund 24000 gas, if a contract is destructed.

The refund happens when value is reset, or contract is destructed. However, note that before a value is reset, it must be set first, and before a contract is destructed, it must be created first. As a result, the full cycle of gas refund goes as follows:

Set-reset cycle: if a storage value is set, charge 20000 gas. If a storage value is reset, charge 5000 gas and refund 15000 gas.
Create-destruct cycle: if a contract is created, charge 32000 gas (together with possible init code execution and deployment gas). If a contract is destructed, refund 24000 gas.

From above, we can see that:

In set-reset cycle, when setting a storage value, it has an absolute cost of at least 10000 gas. The cost will be 10000 more if the value is not reset.
In create-destruct cycle, when creating a contract, it has an absolute cost of at least 8000 gas. The cost will be 24000 more if the contract is not later destructed.

Absolute cost

This allows us to draw the absolute cost table:

Operation	Absolute Gas Cost
Set and reset	10000
Set but not reset	20000
Create and destruct	8000
Create but not destruct	32000

The absolute cost marks the actual utility of the state growth, relative to other opcode’s gas cost. If the absolute cost is increased, then bloating the state will be more expensive, and thus slow down state growth, and vice versa.

Refund is a way to incentivize resetting storage values, and destructing contracts, to free space in the state. The machanic contributes to the calculation of absolute cost table, but is not the source. If refund is removed, the new absolute cost table will be as follows:

Operation	Absolute Gas Cost
Set and reset	25000
Set but not reset	20000
Create and destruct	3200
Create but not destruct	32000

Note that because resetting storage charges an additional 5000 gas, not resetting the value now even has cheaper absolute gas cost compared with resetting the value. The gas scheme certainly becomes less useful without refund.

Instead of removing the refund, what people might actually want is to increase the gas cost of state growth related opcodes. For example, if SSTORE set is increased to 30000 gas, and CREATE initial cost is increased to 56000 gas, we will have the following absolute gas cost table.

Operation	Absolute Gas Cost
Set and reset	20000
Set but not reset	30000
Create and destruct	32000
Create but not destruct	56000

Work around gas refund

The utility GasToken provides can be stated as follows — spend some gases, you can save a portion of them now, and spend them later when gas price increases. This utility still preserves even when gas refund mechanic is removed.

To provide an example, consider using the storage value set-reset cycle, we can imagine a contract, storing users' balance in a simple array of $index → balance$. Whenever a new user is registered, it pushes a value into the end of the array. If the contract anticipates that the gas price will increase in the future, it can pre-allocate indexes now, by registering non-existant users, and setting their balances to a special value marked as $null$ (for example, 0xffff.ffff). This does not rely on the refund mechanic, but accomplishes the same utility as GasToken.