Scenario tests measuring pre-commit gas cost, by batch size.

Adds a scenario test that measures the gas cost of PreCommitSectorBatch for various batch sizes. When pre-committing 1000 sectors, at the maximum batch size, the per-sector cost is reduced to 15% vs batch size of 1 (so 6.5x improvement). With smaller sector counts the relative improvement of large batches is even better (because the HAMT stays smaller).

Sector count    1,000
Batch size  IPLD cost   Calls cost  Total per sector    Total per batch
1   9,670,098,646   29,233,000  9,699,331   9,699,331
2   5,754,837,950   14,616,500  5,769,454   11,538,908
4   3,735,873,808   7,308,250   3,743,182   14,972,728
8   2,672,493,076   3,654,125   2,676,147   21,409,176
16  2,002,461,804   1,841,679   2,004,303   32,068,848
32  1,487,183,156   935,456 1,488,118   47,619,776

Note that this underestimates the improvement to the current mainnet code, because the single-sector case enjoyed some state access optimisations during implementation in terms of the new PreCommitSectorBatch.

The earlier commits in this PR implement some clean-up and I was getting my head around the VM stats. Feel free to ask for reversion of or push a commit reverting anything that doesn't work for you. The TestMeasurePoRepGas output now looks like

--------------------- Batch 0 ---------------------
fil/5/storageminer:7: calls: 200  gets: 2,984  puts: 1,385  read: 841,713  written: 840,025  avg gets: 14.92, avg puts: 6.93
fil/5/storageminer:7: ipld gas: 1,432,991,578 call gas: 5,846,600
  fil/5/storagemarket:8: calls: 200  gets: 600  puts: 200  read: 168,091  written: 168,600  avg gets: 3.00, avg puts: 1.00
  fil/5/storagemarket:8: ipld gas: 281,139,200 call gas: 5,846,600
  fil/5/storagepower:8: calls: 200  gets: 1,369  puts: 985  read: 501,814  written: 502,825  avg gets: 6.84, avg puts: 4.93
  fil/5/storagepower:8: ipld gas: 839,487,748 call gas: 5,846,600
fil/5/cron:2: calls: 1  gets: 182  puts: 80  read: 181,650  written: 83,922  avg gets: 182.00, avg puts: 80.00
fil/5/cron:2: ipld gas: 129,518,244 call gas: 29,233
  fil/5/storagepower:5: calls: 1  gets: 172  puts: 71  read: 181,220  written: 81,930  avg gets: 172.00, avg puts: 71.00
  fil/5/storagepower:5: ipld gas: 125,419,594 call gas: 29,233
    fil/5/storageminer:17: calls: 1  gets: 148  puts: 64  read: 153,041  written: 79,579  avg gets: 148.00, avg puts: 64.00
    fil/5/storageminer:17: ipld gas: 119,968,996 call gas: 29,233
      fil/5/reward:3: calls: 1  gets: 2  puts: 1  read: 175  written: 843  avg gets: 2.00, avg puts: 1.00
      fil/5/reward:3: ipld gas: 1,330,454 call gas: 29,233
      fil/5/storagepower:9: calls: 1  gets: 2  puts: 1  read: 208  written: 843  avg gets: 2.00, avg puts: 1.00
      fil/5/storagepower:9: ipld gas: 1,330,454 call gas: 29,233
      fil/5/storagepower:6: calls: 1  gets: 3  puts: 2  read: 221  written: 1,005  avg gets: 3.00, avg puts: 2.00
      fil/5/storagepower:6: ipld gas: 1,700,366 call gas: 29,233
    fil/5/reward:4: calls: 1  gets: 2  puts: 2  read: 175  written: 962  avg gets: 2.00, avg puts: 2.00
    fil/5/reward:4: ipld gas: 1,569,224 call gas: 29,233
  fil/5/storagemarket:9: calls: 1  gets: 8  puts: 8  read: 362  written: 1,149  avg gets: 8.00, avg puts: 8.00
  fil/5/storagemarket:9: ipld gas: 2,768,196 call gas: 29,233

filecoin-project / specs-actors

Scenario tests measuring pre-commit gas cost, by batch size. #1426

Codecov Report