Open jakmeier opened 1 year ago
FYI, memory cost estimations may be further affected by https://github.com/near/nearcore/pull/8276
CE was broken for a while, the changes we see since the last successful run look like they are due #8224 and #8276
Report
Status: Warn
Current commit: 523f195ac8775414d1f95602808fc5512ca6d91b
Compared to: 4f8a2b4314b08307a0bcb2f4cfa99bd774726ac1
Relative gas estimation changes above threshold: 10
ActionDeleteKey 15.60 Ggas ➜ 18.72 Ggas (+20.00%)
ActionDeployContractPerByte 257.50 Mgas ➜ 231.62 Mgas (-10.05%)
ActionFunctionCallBase 11.16 Ggas ➜ 12.96 Ggas (+16.16%)
ContractCompileBytes 239.38 Mgas ➜ 214.69 Mgas (-10.31%)
ContractCompileBytesV2 233.14 Mgas ➜ 208.18 Mgas (-10.71%)
DataReceiptCreationBase 35.13 Ggas ➜ 38.94 Ggas (+10.86%)
ReadMemoryBase 198.06 Mgas ➜ 166.75 Mgas (-15.81%)
ReadMemoryByte 39399 gas ➜ 25711 gas (-34.74%)
WriteRegisterBase 291.15 Mgas ➜ 259.53 Mgas (-10.86%)
WriteRegisterByte 39497 gas ➜ 25783 gas (-34.72%)
Our gas cost estimations for writing memory were designed before the optimization in https://github.com/near/nearcore/pull/8224. Before vs after the change also showed a significant change in costs in our daily qemu estimation.
It seems that only
WriteMemoryByte
was really affected by it.Here are the most similar costs, with basically no changes:
Qemu
Time (just fro completeness, these numbers have a high variation and cannot be trusted too much)
TODO
The estimation should be updated, as it no longer hits the worst case. See also this comment: https://github.com/near/nearcore/pull/8224#discussion_r1049678652