https://docs.google.com/document/d/186S5yqQ-8HA9KR18U3slS_V1gPx8NjVC3y-ny4aaBnI/edit#heading=h.hlbzjvjqfinf

• New IO gas benchmarks
  • Review and merge !9991 (merged)
    • Manage to run new benchmarks locally
    • Manage to run new benchmarks on a benchmark machine
    • Reproduce previous observations: RAID0 vs RAID5
  • Update biweekly benchmark runs to include the new benchmarks
    • The new benchmarks need specific files: update the script to have them available Very tedious to put in place, with little benefits.
    • The new benchmarks need the RAID0 machine. Either:
      • Run only the IO benchmarks on the RAID0 machine !14917 (merged); or
      • Run all the benchmarks on the RAID0 machine Can be done later anyways
    • Update codegen and gas values in the json (do not merge new values yet, only make sure it's possible and reproducible)
• Update Gas values on mainnet
  • Create feature flag for alternative gas values
  • Either (to be decided):
    • Use new values only for new contracts; or
    • Apply new values to all contracts. Next step becomes then mandatory
  • List all mainnet contacts affected by the gas changes
• Improve gas monitoring
  • Move gas reports from AWS to GCP
    • Put new benchmarks in GCP !14562 (merged)
    • Transfer old benchmarks from AWS to GCP
    • Stop using AWS
  • Fix Michelson generation getting SIGKILLed !14821 (merged)
  • Make sure all resources are
    • Easy to access
    • Up to date
    • With steps that are easy to follow and reproduce
  • Onboard the L1 team on Gas
  • Improve visualization of the evolution of gas
    • Provide more information on gas benchmarks themselves
    • Have inference graphs accessible
    • Have access to the history of a given benchmark/variable
    • (if possible) Provide pointers to the carbonated code for a given benchmark