[single-node-perf] Adjust calibration for V2 loader

[georgemitenkov]

Description

Recalibrated TPS with V2 loader (and first-access validation of global cache accesses). Results:

• Large module working set: clear performance improvement
• Module publishing: clear performance improvement
• Single module: a slight regression within acceptable range, expected because for each cached module we access state view at least once per block (unlike in V2). We can improve these numbers in the future.

How Has This Been Tested?

Key Areas to Review

Type of Change

• New feature
• Bug fix
• Breaking change
• Performance improvement
• Refactoring
• Dependency update
• Documentation update
• Tests

Which Components or Systems Does This Change Impact?

• Validator Node
• Full Node (API, Indexer, etc.)
• Move/Aptos Virtual Machine
• Aptos Framework
• Aptos CLI/SDK
• Developer Infrastructure
• Move Compiler
• Other (specify)

Checklist

• I have read and followed the CONTRIBUTING doc
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I identified and added all stakeholders and component owners affected by this change as reviewers
• I tested both happy and unhappy path of the functionality
• I have made corresponding changes to the documentation

[trunk-io]

⏱️ 1h 46m total CI duration on this PR

Slowest 15 Jobs	Cumulative Duration	Recent Runs
execution-performance / single-node-performance	1h 8m	🟥 🟥
rust-move-tests	9m	🟩
rust-move-tests	9m	🟩
execution-performance / test-target-determinator	7m	🟩 🟩
test-target-determinator	3m	🟩
rust-cargo-deny	3m	🟩 🟩
check-dynamic-deps	3m	🟩 🟩 🟩
semgrep/ci	1m	🟩 🟩 🟩
general-lints	55s	🟩 🟩
file_change_determinator	24s	🟩 🟩
Backport PR	17s	🟥 🟥 🟥
permission-check	14s	🟩 🟩 🟩
file_change_determinator	10s	🟩
permission-check	8s	🟩 🟩 🟩
permission-check	8s	🟩 🟩 🟩

🚨 2 jobs on the last run were significantly faster/slower than expected

Job	Duration	vs 7d avg	Delta
execution-performance / single-node-performance	34m	15m
test-target-determinator	3m	4m

_{settings ⋅ feedback ⋅ docs ⋅ learn more about trunk.io}

[georgemitenkov]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• [single-node-perf] Adjust calibration for V2 loader #15175 👈
• [loader] Generation counters to keep global module cache in sync #15167
• [move] Enable V2 loader #15155 : 1 other dependent PR (#15168 )
• [loader] Global caches for execution #15192 : 2 other dependent PRs (#15255 , #15256 )
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @georgemitenkov and the rest of your teammates on Graphite

[gelash]

As discussed, current implementation seems equivalent to not caching across the block - a big dependency tree benchmark that we want to add later would probably show much more significant degradation.
Let's try to see if we can solve it by comparing parent hashes, some similar idea to here https://github.com/aptos-labs/aptos-core/pull/7873/files#diff-a5693c2ce7821b0a6fec832f8a149a38ceb78901ba36d729c1235a2a9aabb88d.