Benchmark EVM execution performance for multiple execution engine with mainnet data

[dajuguan]

Motivation

Go’s scheduler delivers only about 1/4 the performance of native execution in CPU-intensive workloads, and cgo calls introduce significant overhead. To address this, we aim to evaluate the performance of the Zig-based client (Guillotine) under the BAL scenario.

Goal

Assess the potential performance improvement from parallel execution using the Zig client.

Tasks

• Run Guillotine on Linux (currently it can only be compiled on Mac machine)
• Extract the pre-state for a 500-block execution
• Load the pre-state and write the post-state to memory in Guillotine
• Profile Guillotine performance with and without parallel execution
• Profile Geth performance with and without parallel execution

[qizhou]

It would be great to put motivation and a goal for the issue. Also, we may break down the task into specific sub-tasks, such as running guillotine in linux and extracting pre-state for 500 block execution.