core/vm: implement contract pool

[MatusKysel]

Optimize Contract Pool Implementation

• Replaced custom contract pool implementation with a global sync.Pool
• I didn't measure it for geth but for bnb chain it improve cpu performance more here vm: implement contract pool bnb-chain/bsc#3156

[rjl493456442]

Deployed on 5 and 6. Let's see the impact of this change first.

[rjl493456442]

[rjl493456442]

The PR is slightly faster. The higher IOWait should be relevant with the pebble change (low the compaction debt), I guess it's not relevant with this PR.

[rjl493456442]

[[ PR ]]

[[ MASTER ]]

No big difference?

I can't reproduce the big difference in the flame graph you mentioned in the BSC PR

[rjl493456442]

[[ PR ]]

[[ MASTER ]]

[MatusKysel]

Hey @rjl493456442 can you check specifically vm.NewContract call in pprof? also it might be less significant for ETH right now as we have higher TPS

[rjl493456442]

[[ PR ]]

[[ MASTER ]]

Not sure how to interpret it. This PR indeed eliminates the overhead on vm.NewContract. But this function is so tiny and has no impact on the overall performance

[[ pr ]]

[[ master ]]

[rjl493456442]

Thanks for contributing back! Unfortunately I think the impact of this optimization is too small and the complexity is not worthwhile. Perhaps in the BSC it's somehow obviously and but impact in Ethereum mainnet is not observable.

I will close it, and appreciate if you have any idea for optimizing Geth further!