Great paper analyzing gas pricing

The most notable things to me from this are:

  • Several opcodes are currently vastly underpriced in gas. Particularly,

    • SDIV
    • SLOAD
    • BALANCE
    • SGT
    • BLOCKHASH is also mentioned but that gas fee went up in Constantinople. It may still need to be increased further.
  • Nodes with beefy RAM seem to have significant competitive advantages over more consumer grade nodes (16GB).

  • There’s likely some database optimizations we could make in LEVELDB to improve performance and rreduce variability—particularly on lower-end machines.

  • It may be sensible for client operators to publish suggested specs for minimal hardware.

  • Verifiable computing methods will definitely help.

I’d like to see similar tests carried out against other client software (notably geth and parity) before any conclusions were drawn against the data.