Skip to content

perf(l2): compute base blob fee once per block#6001

Closed
xqft wants to merge 6 commits into
zerocopy_triefrom
opt/p2-fake-exp
Closed

perf(l2): compute base blob fee once per block#6001
xqft wants to merge 6 commits into
zerocopy_triefrom
opt/p2-fake-exp

Conversation

@xqft
Copy link
Copy Markdown
Contributor

@xqft xqft commented Jan 22, 2026 •
edited
Loading

Summary

Optimizes EIP-4844 blob gas calculation by computing base_blob_fee_per_gas once per block instead of twice per transaction, eliminating redundant fake_exponential calls per block.

Motivation

The base blob fee per gas is constant for all transactions in a block (determined by parent_excess_blob_gas and base_fee_update_fraction), yet it was being recomputed:

  • ~768 times in setup_env() (once per transaction)
  • ~768 times in calculate_blob_gas_cost() during transaction execution (via deduct_caller)

This optimization computes the value once at the block level and reuses it, while also simplifying the codebase by removing conditional compilation that was added in a previous approach.

Changes

Removed:

  • fake_exponential_i128() function and all zkVM-specific conditional compilation (#[cfg(any(feature = "zisk", feature = "sp1", feature = "risc0"))])
  • 89 lines of code across block.rs and utils.rs

Modified:

  • calculate_blob_gas_cost() now accepts pre-computed base_blob_fee_per_gas from Environment
  • execute_block() and execute_block_pipeline() compute the value once before the transaction loop
  • setup_env() accepts base_blob_fee_per_gas as a parameter instead of computing it
  • Updated all call sites: execute_tx, execute_tx_in_block, trace_tx_calls, warm_block, rerun_block

Performance Results

ZisK Profile - Block 24291039 (mainnet, 382 transactions)

Metric Baseline Optimized Change
Total Steps 641,671,652 591,264,620 -50.4M (-7.86%)
Total Cost 72.8B 68.1B -4.7B (-6.42%)

Function-Level Impact

Function Baseline Optimized Improvement
LEVM::execute_tx (382 calls) 38.9B 34.3B -11.96%
LEVM::execute_block 41.2B 36.6B -11.33%
VM::execute (386 calls) 35.1B 32.8B -6.70%

Memory Operations

  • memcpy calls: 921,521 → 832,937 (-88,584, -9.6%)
  • memcpy cost: 13.5B → 12.6B (-900M, -6.7%)

Benefits

  1. Performance: Major reduction in zkVM steps/cost
  2. Simplicity: Net -62 lines, removes all conditional compilation
  3. Maintainability: Single code path for all execution contexts
  4. Architecture: Block-level constants computed at block level (correct design)

Test Plan

  • ✅ cargo test -p ethrex-common - All tests pass
  • ✅ cargo test -p ethrex-vm - All tests pass
  • ✅ cargo test -p ethrex-levm - All tests pass
  • ✅ ZisK profile: -7.86% steps on mainnet block

Related

  • Part of zkVM optimization workflow
  • Tracked in scripts/zkvm-bench/logbook.md (Entry docs: add milestones #1)
  • Points to: zerocopy_trie base branch

xqft added 3 commits January 22, 2026 12:12
@xqft
perf(block): add i128 fast path for fake_exponential in zkVM contexts
00b1c47 
Replace U256 arithmetic with i128 for EIP-4844 blob gas calculation when
running in zkVM (zisk/sp1/risc0). This is safe because:
- factor = MIN_BASE_FEE_PER_BLOB_GAS (1)
- numerator = parent_excess_blob_gas (u64, typically < 100M)
- denominator = update_fraction (u64, ~8.8M for mainnet)

Expected savings: 25-30M steps (3.8-4.5%) from reduced checked arithmetic
overhead and smaller integer operations.

Includes test to verify i128 matches U256 for typical values.
@xqft
feat(common): add zisk/sp1 feature flags for zkVM optimizations
a683dd3
@xqft
perf(levm): use i128 fake_exponential in get_base_fee_per_blob_gas fo…
38c1aae 
…r zkVM

Apply same i128 optimization to levm's get_base_fee_per_blob_gas function
which is called during transaction execution. This is the main call site
with ~769 calls per block.
@github-actions github-actions Bot added the L2 Rollup client label Jan 22, 2026
@github-actions
Copy link
Copy Markdown

github-actions Bot commented Jan 22, 2026 •
edited
Loading

Lines of code report

Total lines added: 32
Total lines removed: 2
Total lines changed: 34

Detailed view 
+-------------------------------------------------+-------+------+
| File                                            | Lines | Diff |
+-------------------------------------------------+-------+------+
| ethrex/crates/vm/backends/levm/mod.rs           | 711   | +17  |
+-------------------------------------------------+-------+------+
| ethrex/crates/vm/backends/levm/tracing.rs       | 71    | +10  |
+-------------------------------------------------+-------+------+
| ethrex/crates/vm/backends/mod.rs                | 176   | +5   |
+-------------------------------------------------+-------+------+
| ethrex/crates/vm/levm/src/hooks/default_hook.rs | 351   | -1   |
+-------------------------------------------------+-------+------+
| ethrex/crates/vm/levm/src/utils.rs              | 431   | -1   |
+-------------------------------------------------+-------+------+

@github-actions
Copy link
Copy Markdown

github-actions Bot commented Jan 22, 2026 •
edited
Loading

Benchmark Results Comparison

No significant difference was registered for any benchmark run.

Detailed Results

Benchmark Results: BubbleSort

Command Mean [s] Min [s] Max [s] Relative
main_revm_BubbleSort 2.995 ± 0.021 2.963 3.028 1.00
main_levm_BubbleSort 3.036 ± 0.039 3.005 3.130 1.01 ± 0.01
pr_revm_BubbleSort 3.007 ± 0.031 2.965 3.059 1.00 ± 0.01
pr_levm_BubbleSort 3.044 ± 0.009 3.032 3.060 1.02 ± 0.01

Benchmark Results: ERC20Approval

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_ERC20Approval 977.8 ± 5.4 970.7 989.3 1.00
main_levm_ERC20Approval 1064.9 ± 19.3 1051.0 1116.1 1.09 ± 0.02
pr_revm_ERC20Approval 988.6 ± 9.2 977.3 1003.2 1.01 ± 0.01
pr_levm_ERC20Approval 1094.8 ± 9.7 1082.5 1112.8 1.12 ± 0.01

Benchmark Results: ERC20Mint

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_ERC20Mint 135.2 ± 1.6 134.2 139.3 1.01 ± 0.01
main_levm_ERC20Mint 159.4 ± 1.5 158.0 162.9 1.20 ± 0.01
pr_revm_ERC20Mint 133.3 ± 0.7 132.3 134.9 1.00
pr_levm_ERC20Mint 160.2 ± 0.8 159.1 161.4 1.20 ± 0.01

Benchmark Results: ERC20Transfer

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_ERC20Transfer 232.4 ± 1.9 230.3 236.3 1.00 ± 0.01
main_levm_ERC20Transfer 268.0 ± 2.5 265.5 273.5 1.15 ± 0.01
pr_revm_ERC20Transfer 232.2 ± 1.6 230.1 235.1 1.00
pr_levm_ERC20Transfer 273.7 ± 4.7 269.6 286.5 1.18 ± 0.02

Benchmark Results: Factorial

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_Factorial 222.9 ± 1.5 221.4 225.6 1.01 ± 0.01
main_levm_Factorial 255.2 ± 1.5 253.6 257.9 1.15 ± 0.01
pr_revm_Factorial 221.5 ± 2.3 219.3 226.5 1.00
pr_levm_Factorial 261.5 ± 16.1 253.0 306.6 1.18 ± 0.07

Benchmark Results: FactorialRecursive

Command Mean [s] Min [s] Max [s] Relative
main_revm_FactorialRecursive 1.641 ± 0.051 1.512 1.688 1.00
main_levm_FactorialRecursive 8.298 ± 0.054 8.169 8.353 5.06 ± 0.16
pr_revm_FactorialRecursive 1.689 ± 0.071 1.612 1.860 1.03 ± 0.05
pr_levm_FactorialRecursive 8.279 ± 0.044 8.192 8.350 5.04 ± 0.16

Benchmark Results: Fibonacci

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_Fibonacci 200.6 ± 1.3 198.4 203.5 1.03 ± 0.01
main_levm_Fibonacci 249.8 ± 4.8 241.4 257.5 1.28 ± 0.03
pr_revm_Fibonacci 195.7 ± 0.9 194.6 197.7 1.00
pr_levm_Fibonacci 248.8 ± 6.3 235.7 260.3 1.27 ± 0.03

Benchmark Results: FibonacciRecursive

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_FibonacciRecursive 873.1 ± 8.6 862.9 888.1 1.24 ± 0.02
main_levm_FibonacciRecursive 706.5 ± 5.8 698.0 714.3 1.00
pr_revm_FibonacciRecursive 870.5 ± 10.6 854.0 892.9 1.23 ± 0.02
pr_levm_FibonacciRecursive 715.2 ± 13.0 699.4 742.6 1.01 ± 0.02

Benchmark Results: ManyHashes

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_ManyHashes 8.4 ± 0.1 8.3 8.5 1.00
main_levm_ManyHashes 9.0 ± 0.1 8.9 9.3 1.08 ± 0.02
pr_revm_ManyHashes 8.5 ± 0.1 8.4 8.9 1.02 ± 0.02
pr_levm_ManyHashes 9.1 ± 0.1 8.9 9.3 1.08 ± 0.02

Benchmark Results: MstoreBench

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_MstoreBench 262.9 ± 6.4 258.4 274.8 1.19 ± 0.03
main_levm_MstoreBench 224.9 ± 9.9 220.5 252.9 1.02 ± 0.05
pr_revm_MstoreBench 256.6 ± 0.9 255.4 258.2 1.16 ± 0.01
pr_levm_MstoreBench 220.6 ± 1.8 219.0 223.6 1.00

Benchmark Results: Push

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_Push 290.3 ± 13.4 284.6 328.3 1.03 ± 0.05
main_levm_Push 283.5 ± 1.4 282.2 286.8 1.01 ± 0.03
pr_revm_Push 284.8 ± 0.6 283.4 285.7 1.01 ± 0.03
pr_levm_Push 281.7 ± 7.1 272.2 298.1 1.00

Benchmark Results: SstoreBench_no_opt

Command Mean [ms] Min [ms] Max [ms] Relative
main_revm_SstoreBench_no_opt 168.9 ± 4.4 162.5 174.6 1.87 ± 0.09
main_levm_SstoreBench_no_opt 91.3 ± 5.6 86.4 104.8 1.01 ± 0.07
pr_revm_SstoreBench_no_opt 172.4 ± 8.0 162.1 185.8 1.91 ± 0.11
pr_levm_SstoreBench_no_opt 90.2 ± 3.4 86.5 96.0 1.00

@xqft @claude
Optimize base blob fee calculation by computing once per block
40885a4 
Replace the i128 conditional compilation approach with block-level caching
to reduce redundant fake_exponential calls from ~1536 per block to 1.

Changes:
- Remove fake_exponential_i128() and all cfg(zkvm) conditional compilation
- Update calculate_blob_gas_cost() to use pre-computed base_blob_fee_per_gas
  from Environment instead of recomputing (eliminates ~768 calls/block)
- Compute base_blob_fee_per_gas once in execute_block/execute_block_pipeline
  and pass to all transactions (eliminates ~768 calls/block)
- Thread the pre-computed value through setup_env, execute_tx, and
  execute_tx_in_block functions
- Update all call sites in tracing, warm_block, and Evm wrapper

Benefits:
- Simpler codebase: removes 89 lines of conditional compilation
- Same or better performance: 1 U256 fake_exponential vs 1536 i128 operations
- More maintainable: single code path for all execution contexts
- Architecturally correct: block-level constants computed at block level

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
@xqft xqft changed the title chore(l2): optimize EIP-4844 blob gas calculation with i128 for zkVM chore(l2): optimize EIP-4844 blob gas by computing once per block Jan 22, 2026
@xqft xqft changed the title chore(l2): optimize EIP-4844 blob gas by computing once per block perf(l2): optimize EIP-4844 blob gas by computing once per block (-7.86%) Jan 22, 2026
@xqft @claude
Add zkVM optimization logbook with P2 block-level caching entry
ef5f483 
Document the block-level base blob fee caching optimization that achieved
-7.86% step reduction (-50.4M steps) on mainnet block 24291039.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
@github-actions github-actions Bot added the performance Block execution throughput and performance in general label Jan 22, 2026
@github-actions
Copy link
Copy Markdown

github-actions Bot commented Jan 22, 2026 •
edited
Loading

Benchmark Block Execution Results Comparison Against Main

Command Mean [s] Min [s] Max [s] Relative
base 65.534 ± 0.473 64.925 66.423 1.00
head 65.717 ± 0.260 65.353 66.052 1.00 ± 0.01

@xqft xqft changed the title perf(l2): optimize EIP-4844 blob gas by computing once per block (-7.86%) perf(l2): optimize EIP-4844 blob gas by computing once per block Jan 22, 2026
@xqft xqft changed the title perf(l2): optimize EIP-4844 blob gas by computing once per block perf(l2): compute blob gas fee once per block Jan 22, 2026
@xqft xqft changed the title perf(l2): compute blob gas fee once per block perf(l2): compute base blob fee once per block Jan 22, 2026
@xqft
Copy link
Copy Markdown
Contributor Author

xqft commented Jan 22, 2026

superseded by #6006

@xqft xqft closed this Jan 22, 2026
@github-project-automation github-project-automation Bot moved this from Todo to Done in ethrex_performance Jan 22, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

@xqft xqft

Labels

L2 Rollup client performance Block execution throughput and performance in general

Projects

ethrex_l2
Status: Done
ethrex_performance
Status: Done

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@xqft