Computation Cost
Since Klaytn aims to maintain 1-second block time, the execution time of transactions has to be managed. Here are three approaches to achieve that:
- Limiting the gas limit of a transaction
- Limiting the execution time of a transaction
- Limiting the computation cost of a transaction
Limiting the gas limit of a transaction was not a feasible solution because the concept of the gas represents the current exchange value of the various resources in the blockchain platform such as computation, storage, network bandwidth, and so on. It is not suitable as a metric for the transaction execution time.
Limiting the execution time of a transaction was not feasible either because the execution time can vary between nodes on the blockchain platform. For example, consider the case in which we limit the execution time of a transaction to be 100 milli-second. If a node executes a transaction in 90 ms and another node executes it in 110 ms, the two nodes cannot reach a consensus. Hence, this solution is not appropriate.
The last approach is to limit the computation cost of a transaction. We modelled the computation cost of each EVM opcode based on its actual execution time and limit the sum of computation cost of a transaction. With this approach, we eliminate other factors and only count the normalized execution time unit, and nodes can reach a consensus as well.
Therefore, we chose the third option for Klaytn. The computation cost limit was 100,000,000, but as CPU computing performance has increased, the limit has been raised to 150,000,000 after Cancun EVM hardfork. This limit value is determined by the platform, so the developers should be aware of the computation cost of a transaction. To calculate the computation cost of a transaction, Klaytn provides klay_estimateComputationCost. The usage is almost the same as klay_estimateGas.
Computation cost related hardfork changes can be found at the bottom of this page. Go to Hardfork Changes.
Computation Cost Limit
A series of opcodes or precompiled contracts are executed sequentially when executing a transaction. To limit the execution time of a transaction, we have made a deterministic execution time calculation model for opcodes and precompiled contracts based on real execution time.
Based on this model, predetermined computation cost values for opcodes and precompiled contracts are added to the total computation cost. If the total value exceeds computation cost limit, transaction execution is aborted and returns ComputationCostLimitReached(0x0a) error.
When setting the computation cost limit value, we set --opcode-computation-cost-limit flag value as a limit if it is set as a non-zero value. If it's zero, the limit is set to the default computation cost limit defined for each specific hardfork.
Exceptionally, the limit for call/estimateGas/estimateComputationCost is always set to unlimited and is not influenced by flag or hardfork values. However, execution still can be aborted due to other limits such as gas cap.
Computation Cost of Opcodes
The below table shows the computation cost of EVM opcodes. The computation cost was determined based on experiments.
| Opcode | ComputationCost |
|---|---|
| STOP | 0 |
| ADD | 150 |
| MUL | 200 |
| SUB | 219 |
| DIV | 404 |
| SDIV | 360 |
| MOD | 320 |
| SMOD | 560 |
| ADDMOD | 360 |
| MULMOD | 700 |
| EXP | 720 |
| SIGNEXTEND | 481 |
| LT | 201 |
| GT | 264 |
| SLT | 176 |
| SGT | 222 |
| EQ | 220 |
| ISZERO | 165 |
| AND | 288 |
| OR | 160 |
| XOR | 454 |
| NOT | 364 |
| BYTE | 589 |
| SHL | 478 |
| SHR | 498 |
| SAR | 834 |
| SHA3 | 560 |
| ADDRESS | 284 |
| BALANCE | 1407 |
| ORIGIN | 210 |
| CALLER | 188 |
| CALLVALUE | 149 |
| CALLDATALOAD | 596 |
| CALLDATASIZE | 194 |
| CALLDATACOPY | 100 |
| CODESIZE | 145 |
| CODECOPY | 898 |
| GASPRICE | 131 |
| EXTCODESIZE | 1481 |
| EXTCODECOPY | 1000 |
| RETURNDATASIZE | 10 |
| RETURNDATACOPY | 40 |
| EXTCODEHASH | 1000 |
| BLOCKHASH | 500 |
| COINBASE | 189 |
| TIMESTAMP | 265 |
| NUMBER | 202 |
| PREVRANDAO | 1498 |
| GASLIMIT | 166 |
| CHAINID | 120 |
| SELFBALANCE | 374 |
| POP | 140 |
| MLOAD | 376 |
| MSTORE | 288 |
| MSTORE8 | 230 |
| SLOAD | 2550 |
| SSTORE | 2510 |
| JUMP | 253 |
| JUMPI | 176 |
| PC | 147 |
| MSIZE | 137 |
| GAS | 230 |
| JUMPDEST | 10 |
| PUSH0 | 80 |
| PUSH1 | 120 |
| PUSH2 | 120 |
| PUSH3 | 120 |
| PUSH4 | 120 |
| PUSH5 | 120 |
| PUSH6 | 120 |
| PUSH7 | 120 |
| PUSH8 | 120 |
| PUSH9 | 120 |
| PUSH10 | 120 |
| PUSH11 | 120 |
| PUSH12 | 120 |
| PUSH13 | 120 |
| PUSH14 | 120 |
| PUSH15 | 120 |
| PUSH16 | 120 |
| PUSH17 | 120 |
| PUSH18 | 120 |
| PUSH19 | 120 |
| PUSH20 | 120 |
| PUSH21 | 120 |
| PUSH22 | 120 |
| PUSH23 | 120 |
| PUSH24 | 120 |
| PUSH25 | 120 |
| PUSH26 | 120 |
| PUSH27 | 120 |
| PUSH28 | 120 |
| PUSH29 | 120 |
| PUSH30 | 120 |
| PUSH31 | 120 |
| PUSH32 | 120 |
| DUP1 | 190 |
| DUP2 | 190 |
| DUP3 | 176 |
| DUP4 | 142 |
| DUP5 | 177 |
| DUP6 | 165 |
| DUP7 | 147 |
| DUP8 | 157 |
| DUP9 | 138 |
| DUP10 | 174 |
| DUP11 | 141 |
| DUP12 | 144 |
| DUP13 | 157 |
| DUP14 | 143 |
| DUP15 | 237 |
| DUP16 | 149 |
| SWAP1 | 141 |
| SWAP2 | 156 |
| SWAP3 | 145 |
| SWAP4 | 135 |
| SWAP5 | 115 |
| SWAP6 | 146 |
| SWAP7 | 199 |
| SWAP8 | 130 |
| SWAP9 | 160 |
| SWAP10 | 134 |
| SWAP11 | 147 |
| SWAP12 | 128 |
| SWAP13 | 121 |
| SWAP14 | 114 |
| SWAP15 | 197 |
| SWAP16 | 128 |
| LOG0 | 100 |
| LOG1 | 500 |
| LOG2 | 500 |
| LOG3 | 500 |
| LOG4 | 500 |
| PUSH | 0 |
| DUP | 0 |
| SWAP | 0 |
| CREATE | 2094 |
| CALL | 5000 |
| CALLCODE | 4000 |
| RETURN | 0 |
| DELEGATECALL | 696 |
| CREATE2 | 10000 |
| STATICCALL | 10000 |
| REVERT | 0 |
| SELFDESTRUCT | 0 |
| BASEFEE | 198 |
| BLOBBASEFEE | 120 |
| BLOBHASH | 165 |
| TSTORE | 280 |
| TLOAD | 220 |
| MCOPY | 250 |
Precompiled contracts computation cost table
Input is a byte array input of a precompiled contract.
| Address | Precompiled contracts | Computation Cost |
|---|---|---|
| 0x01 | ecrecover | 113,150 |
| 0x02 | sha256hash | numOfWords(input) / 32 * 100 + 1,000 |
| 0x03 | ripemd160hash | numOfWords(input) / 32 * 10 + 100 |
| 0x04 | dataCopy | 0 |
| 0x05 | bigModExp | see the code here |
| 0x06 | bn256Add | 8,000 |
| 0x07 | bn256ScalarMul | 100,000 |
| 0x08 | bn256Pairing | numOfPairings(input) * 1,000,000 + 2,000,000 |
| 0x09 | blake2f | bigEndian(getRounds(input[0:4])) * 10 + 10,000 |
| 0x0A | kzg | 2,200,000 |
| 0x3FD | vmLog | 10 |
| 0x3FE | feePayer | 10 |
| 0x3FF | validateSender | numOfSigs(input) * 180,000 + 10,000 |
Hardfork Changes
| Hardfork | New items | Changes | ||
|---|---|---|---|---|
| Cancun EVM | BLOBBASEFEE (0x49) BLOBHASH (0x50) TSTORE (0x5c) opcode TLOAD (0x5d) MCOPY (0x5e) kzg (0x0a) precompiled contract | increase the computation cost limit from 100,000,000 to 150,000,000 reduce the computation cost of some opcodes due to cpu performance increase -Sdiv (0x05): 739 -> 360 -Mod (0x06): 812 -> 320 -Addmod (0x08): 1410 -> 360 -Mulmod (0x09): 1760 -> 700 -Exp (0x0A): 5000 -> 720 -Sha3 (0x20): 2465 -> 560 -Mstore8 (0x53): 5142 -> 230 -Log1, Log2, Log3, Log4 (0xA1-0xA4): 1000 -> 500 increase the computation cost of some opcodes due to increased database size -SLOAD (0x54): 835 -> 2550 -SSTORE (0x55): 1548 -> 2510 | ||
| Shanghai EVM | PUSH0 (0x5f) opcode | |||
| Kore | modExp (0x05) precompiled contract use new gas calculation logic. Computation cost also affected. Become more accurate. | |||
| London EVM | BaseFee (0x48) opcode | |||
| Istanbul EVM | CHAINID (0x46) opcode SELFBALANCE (0x47) opcode blake2f (0x09) precompiled contract | reduce the computation cost of over-priced opcodes - ADDMOD (0x08): 3349 -> 1410 - MULMOD (0x09): 4757 -> 1760 - XOR (0x18): 657 -> 454 - NOT (0x19): 1289 -> 364 - SHL (0x1B): 1603 -> 478 - SHR (0x1C): 1815 -> 834 |