Skip to content

BENCHMARKS.md

Latest commit

 

History

History
197 lines (157 loc) · 7.36 KB

BENCHMARKS.md

File metadata and controls

197 lines (157 loc) · 7.36 KB
Raw

Quasar Consensus — Measured Benchmarks

Hardware: Apple M1 Max (10 cores, darwin/arm64) Build: Go 1.26.1, CPU path only (no -tags accel / no CGO GPU)

All numbers below are raw go test -bench output. Reproduce with the commands at the end of each section.

1. BLS12-381 (crypto/bls)

pkg: github.com/luxfi/crypto/bls
BenchmarkNewSecretKey-10           1000000       1056 ns/op
BenchmarkSign-10                      3592     350089 ns/op
BenchmarkVerify-10                    1563     819843 ns/op
BenchmarkAggregatePublicKeys-10       2719     481776 ns/op
BenchmarkAggregateSignatures-10       2306     542316 ns/op

Reproduce:

cd ~/work/lux/crypto/bls && go test -bench=. -benchtime=1s -run=^$

2. ML-DSA-65 (crypto/mldsa)

pkg: github.com/luxfi/crypto/mldsa
BenchmarkMLDSA_Sign-10             5208    494486 ns/op
BenchmarkMLDSA_Verify-10          14754    180723 ns/op
BenchmarkMLDSA_KeyGeneration-10   10000    202173 ns/op

Reproduce:

cd ~/work/lux/crypto && go test -bench=. -benchtime=2s -run=^$ ./mldsa/...

3. UTXO Fx Plugins (with LRU verify cache)

pkg: github.com/luxfi/utxo/mldsafx
BenchmarkMLDSA65Verify-10          9712    253761 ns/op
BenchmarkMLDSA65VerifyCached-10  880334      2963 ns/op

pkg: github.com/luxfi/utxo/slhdsafx
BenchmarkSLH192fVerify-10          2762   1916073 ns/op
BenchmarkSLH192fVerifyCached-10   32167    131072 ns/op

pkg: github.com/luxfi/utxo/ed25519fx
BenchmarkEd25519Verify-10         10000    205551 ns/op
BenchmarkEd25519VerifyCached-10 2289220      1144 ns/op

pkg: github.com/luxfi/utxo/secp256r1fx
BenchmarkP256Verify-10            21979    121321 ns/op
BenchmarkP256VerifyCached-10    2347353      1032 ns/op

pkg: github.com/luxfi/utxo/secp256k1fx
BenchmarkSecp256k1Verify-10     4237567       658.1 ns/op

Reproduce:

cd ~/work/lux/utxo && go test -bench=. -benchtime=2s -run=^$ \
  ./mldsafx/... ./slhdsafx/... ./ed25519fx/... ./secp256r1fx/... ./secp256k1fx/...

4. ML-DSA EVM Precompile

pkg: github.com/luxfi/precompile/mldsa
BenchmarkMLDSAVerify_SmallMessage-10           10000   247749 ns/op
BenchmarkMLDSAVerify_LargeMessage-10            8485   274320 ns/op
BenchmarkMLDSAVerify_AllModes/ML-DSA-44-10     15764   140178 ns/op
BenchmarkMLDSAVerify_AllModes/ML-DSA-65-10     10000   250112 ns/op
BenchmarkMLDSAVerify_AllModes/ML-DSA-87-10      6062   419649 ns/op

Reproduce:

cd ~/work/lux/precompile && go test -bench=. -benchtime=2s -run=^$ ./mldsa/...

5. Quasar Protocol (consensus/protocol/quasar)

pkg: github.com/luxfi/consensus/protocol/quasar
BenchmarkQuasarSigVariableMessageSize/verify_msg_512_bytes-10    1172  1017823 ns/op
BenchmarkQuasarSigVariableMessageSize/sign_msg_1024_bytes-10     1460   808500 ns/op
BenchmarkQuasarSigVariableMessageSize/verify_msg_1024_bytes-10    949  1089206 ns/op
BenchmarkBLSAggregation/4_signatures-10                          5877   209808 ns/op
BenchmarkBLSAggregation/8_signatures-10                          2966   409945 ns/op
BenchmarkBLSAggregation/16_signatures-10                         1406   832009 ns/op
BenchmarkBLSAggregation/32_signatures-10                          724  1656841 ns/op
BenchmarkBLSAggregation/64_signatures-10                          369  3288973 ns/op
BenchmarkBLSAggregation/100_signatures-10                         232  5297844 ns/op
BenchmarkBLSAggregatedVerification/4_signers-10                  1579   820998 ns/op
BenchmarkBLSAggregatedVerification/8_signers-10                  1294   815583 ns/op
BenchmarkBLSAggregatedVerification/16_signers-10                 1509   876531 ns/op
BenchmarkBLSAggregatedVerification/32_signers-10                 1506   832595 ns/op
BenchmarkBLSAggregatedVerification/64_signers-10                 1390   776815 ns/op
BenchmarkBLSAggregatedVerification/100_signers-10                1609   875227 ns/op
BenchmarkQuasarBlockProcessing-10                                 674  1849772 ns/op
BenchmarkQuasarQuantumHash-10                                 2884668      434.7 ns/op
BenchmarkQuasarValidatorAddition-10                              3861   327629 ns/op

Reproduce:

cd ~/work/lux/consensus && GOWORK=off go test \
  -bench='^BenchmarkQuasar|^BenchmarkBLS' \
  -benchtime=1s -run=^$ -timeout=120s ./protocol/quasar/

6. GPU Primitives (Metal, CPU fallback for small N)

pkg: github.com/luxfi/gpu (Metal backend)
BenchmarkMatMul-10     3151   399456 ns/op   20027.26 MB/s
BenchmarkAdd-10        4346   335961 ns/op     238.12 MB/s
  Backend: metal, Device: Apple M1 Max (metal)

pkg: github.com/luxfi/accel/ops/zk (CPU fallback, N=8)
BenchmarkNTT_N8-10        5022608      460.9 ns/op
BenchmarkPolyMul_N8-10    1606084     1546   ns/op
BenchmarkFieldMul-10      1000000     2248   ns/op

pkg: github.com/luxfi/accel (no-CGO default path)
BenchmarkNoCGOPath/DefaultSession_Stub-10  516712508    2.446 ns/op
BenchmarkNoCGOPath/Available_Check-10       44443218   27.03  ns/op
BenchmarkNoCGOPath/Backends_List-10         40341445   29.89  ns/op

Note: accel.BLSBatchVerifyThreshold = 64. Below 64 sigs, CPU is faster due to ~100 μs Metal dispatch overhead. Above 64, GPU batch verify benefits.

Reproduce:

cd ~/work/lux/gpu && go test -bench=. -benchtime=1s -run=^$ .
cd ~/work/lux/accel && go test -bench=. -benchtime=2s -run=^$ ./ops/zk/...
cd ~/work/lux/accel && go test -bench=. -benchtime=1s -run=^$ .

7. EVM (evmgpu core, CPU only)

pkg: github.com/luxfi/evmgpu/core
BenchmarkInsertChain_empty_memdb-10      8019    171136 ns/op  163423 B/op  602 allocs/op
BenchmarkInsertChain_valueTx_memdb-10    5487    245878 ns/op   75289 B/op  881 allocs/op

Translated: 5844 empty blocks/sec; 4067 value-tx blocks/sec.

Reproduce (working benches only):

cd ~/work/lux/evmgpu && go test \
  -bench='BenchmarkInsertChain_empty_memdb|BenchmarkInsertChain_valueTx_memdb' \
  -benchtime=1s -run=^$ -timeout=120s ./core/

ring-call benchmarks currently crash with a pre-existing nil-pointer in core/types.Header.Hash at core/bench_test.go:306 — unrelated to consensus.

8. Derived QuasarCert Numbers

Per-cert production (n=21 validators, 1 cert/epoch):

Component Ops CPU Cost
Per-validator BLS sign 21 × 350 us 7.35 ms total (parallel: 350 us wall)
BLS aggregation 1 × (21 × 53 us) 1.11 ms
Per-validator ML-DSA sign 21 × 495 us 10.4 ms total (parallel: 495 us wall)
Z-Chain Groth16 prover 1 × ~400 ms 400 ms CPU / 5-15 ms GPU (est)
Pulsar round 1 × variable ~ms range
Total wall-clock ~400-500 ms CPU (Groth16-dominated)

Per-cert verification (light client):

Component Cost
BLS aggregated verify (100 signers) 875 us (constant)
Groth16 verify (3 pairings on BLS12-381) ~1-3 ms CPU, ~200-500 us GPU
Pulsar verify variable
Total ~2-5 ms CPU

This supersedes the 357 us claim in older papers. Stale sections to correct:

  • papers/lp-105-quasar-consensus/sections/05-chain-separation-for-threshold-cryptogra.tex:92
  • papers/lp-105-quasar-consensus/sections/07-security-analysis.tex:264
  • papers/lux-triple-proof-consensus/sections/*
  • papers/lux-quasar-consensus/sections/*
  • papers/lux-performance-security-tradeoffs/sections/*
  • papers/lux-master-security-model/sections/*

See ~/work/lux/proofs/quasar-cert-soundness.tex Appendix B for the ML-DSA-65 Groth16 circuit analysis (~2^22.5 R1CS constraints per verification).