Performance

Proven with a real load test, not promised in a sales deck.

543requests/second

185msp95 latency

0%errors

To establish a worst-case performance floor, we load-tested Inzinx on the smallest cloud instance available — and even there, it sustained the numbers above, over real TLS, in a real test against the live API.

Held steady under 5x heavier load

Throughput barely moved and errors stayed at zero when concurrency jumped from 100 to 500 requests at once — latency grew, as expected, but nothing broke.

Everyday load

100 concurrent requests

Throughput710 req/s

p95 latency139ms

0% errors

5x heavier stress test

500 concurrent requests

Throughput680 req/s

p95 latency745ms

0% errors

Why the headroom isn't an accident

Stateless by design

Every eligibility and rating call is a self-contained computation -- no server-side session ties a request to a specific machine. That's what makes scaling straightforward rather than risky.

Kubernetes-ready, already built

Deployment manifests (Deployment, Service, Ingress, autoscaling) are committed and ready to apply -- not a future redesign, just infrastructure waiting on real scale to justify it.

Scales both directions

Vertically with a larger instance, or horizontally with more replicas behind a load balancer -- the architecture supports either path as load grows.

Methodology

Tested with k6 against the live production API (api.inzinx.com), over real TLS, using a dedicated test tenant so no real customer data or traffic was touched. Both scenarios were closed-loop (each simulated client fires its next request immediately after the previous one completes, no think-time) — a deliberate worst case, not an average. Questions about the raw results? Ask us.

See it on your own workload

Every account gets 3 months free with a daily quota — no credit card required.

Get started free