M0.1 Benchmark: WASM-SIMD secp256k1 verify throughput in Web Workers

[melvincarvalho]

Part of #1 (M0). This is the single most important number in the project.

Goal

Measure ECDSA and Schnorr verifications/sec using a WASM-compiled libsecp256k1, single-core and scaled across N Web Workers, on real signatures pulled from mainnet blocks.

Why

Whether full validation is hours or days hinges on this. The engine's current crypto is pure JS (~20-100x too slow); the node needs WASM secp. Everything downstream is speculative until this is measured.

Benchmark / acceptance

• Report verifies/sec/core (ECDSA and Schnorr), with and without WASM SIMD.
• Report total throughput scaling 1..N workers (near-linear?).
• Report the x-factor vs native libsecp on the same machine.
• Use real signatures from downloaded blocks, not synthetic.
• Confirm: >= ~10k verifies/sec/core in WASM, scaling roughly with core count.

Notes

Use an existing WASM build of libsecp256k1. Output feeds M0.5's revised time model.

[melvincarvalho]

Measured (single core, this dev machine), via npm run bench:secp (bench/secp-verify.js):

	ECDSA verify/s	Schnorr verify/s
pure JS (engine secp256k1.js)	~248	~248
WASM (tiny-secp256k1)	~3,840	~3,810

WASM is 16x the pure-JS engine. Projected full-chain signature work (~2.5B ECDSA verifications):

• pure JS, 1 core: ~117 days (the engine's BigInt secp is fine for the test suite, hopeless for IBD)
• WASM, 1 core: ~180 h
• WASM, 8 cores: ~23 h
• WASM, 24 cores: ~7.5 h

This revises the roadmap model down. I had estimated ~10k verify/s/core and ~3h on 24 cores; the real tiny-secp256k1 figure is ~3,840/s. The gap is per-call JS↔WASM marshalling overhead, which makes this WASM ~5-10x off native (not the 1.5-2x kernel figure).

Conclusions:

• The current pure-JS engine cannot do IBD; a WASM secp is mandatory (confirmed).
• The real levers are batch verification (amortize the WASM boundary across many sigs) and a SIMD libsecp build — not just "compile to WASM". tiny-secp256k1 does neither, so it is a floor, not a ceiling.
• Honest full-validation target with this class of WASM: overnight (~8-15h) on a high-core desktop, not "an afternoon (~3h)". Batch+SIMD could pull it back toward the afternoon; that needs its own benchmark.

Next: benchmark a batched / SIMD secp before committing to block validation.