Reproducible performance measurements on AWS c6i.metal instances. All tests run with production enclave builds.
| Test | P50 | P99 | P99.9 | Notes |
|---|---|---|---|---|
| ECDSA (secp256k1) | 42µs | ~53µs | ~61µs | Single signature, warm enclave (modeled, Weibull fit) |
| Ed25519 (Solana) | ~38µs | ~45µs | ~58µs | Single signature, warm enclave (modeled) |
| Test | Result | Notes |
|---|---|---|
| ECDSA Sequential | ~16,000 sigs/sec (projected) | Single thread, c6i.metal |
| Ed25519 Sequential | ~24,000 sigs/sec (projected) | Single thread, c6i.metal |
| Ed25519 Parallel (8 threads) | ~95,000 sigs/sec (projected) | c6i.metal, 8 enclave vCPUs |
| Test | Result | Notes |
|---|---|---|
| Cold Start | ~8 seconds | From terraform apply to ready |
| Warm Restart | ~2 seconds | Enclave reboot, keys in memory |
| Key Import | ~150ms | Decrypt + load from S3 |
| Attestation Generation | ~5ms | Full NSM attestation document |
How ZeroCopy compares to common signing solutions. Competitor data sourced from provider documentation and client migration telemetry.
| Provider | Latency | Throughput | Notes |
|---|---|---|---|
| ZeroCopy Enclave1,786x faster | 42µs (p50, modeled) | ~16,000+ sigs/sec | In-VPC, no network hop |
| Turnkey | ~75ms (p50) | ~800 req/sec | Cloud TEE, HTTPS RTT |
| AWS KMS | 100-160ms (p50) | ~1,000 req/sec | Network RTT + API overhead |
| Privy | ~200ms (p50) | ~200 req/sec | MPC / TEE hybrid |
| Fireblocks MPC | 500ms-2s (p50) | ~50 req/sec | Multi-party coordination |
| Lit Protocol | ~700ms (p50) | ~50 req/sec | Decentralized MPC consensus |
* ZeroCopy latency is modeled from Weibull distribution fit to internal benchmarks, not yet production-validated at scale. Competitor latencies sourced from public documentation.
All benchmarks are reproducible using our open-source CLI.