Performance

Measured with the benchmark workflow on a GitHub-hosted ubuntu-24.04 runner (4 vCPU AMD EPYC 9V45): Corvus pinned to 2 vCPU, the load generator and the backend on the other two, 64 keep-alive connections, GET / with a 3-byte response.

Scenario	req/s	p50	p99	Memory
nginx 1.28 reverse proxy (baseline)	97,227	0.63 ms	1.16 ms	17 MB
Corvus, default host	75,473	0.79 ms	1.83 ms	72 MB
Corvus, 30 rules + 1,000-CIDR list + exploit blocking + headers + rate limit	54,585	1.12 ms	2.29 ms	77 MB
Corvus, HTTPS (ECDSA P-256), keep-alive	59,697	1.00 ms	2.18 ms	80 MB
Corvus, HTTPS, new TLS handshake per request	4,348 handshakes/s	5.88 ms	11.62 ms	88 MB
Corvus, access queue (admitted visitors)	57,954	1.05 ms	2.22 ms	84 MB
Corvus, challenge rule (visitors with a pass)	62,901	0.96 ms	1.94 ms	88 MB

Corvus reaches about three quarters of plain nginx's throughput; the full protection set costs about 28 % on top of that. The runner is shared and the responses are tiny, so the numbers are meant for comparing scenarios and versions, not as capacity guarantees. Memory includes the control plane and the built-in access-queue store.