What is RPS in load testing?

RPS means requests per second completed against the target. It measures throughput for HTTP/API tests and should be read with latency and error rate.

What is the difference between concurrent connections and RPS?

Concurrent connections are simultaneous open sessions or workers. RPS is completed requests per second. High concurrency with low RPS often means the system is waiting or saturating.

What is an abort criterion?

A pre-agreed stop rule — such as p95 latency, error rate, CPU, or NIC saturation — that ends a stress test before uncontrolled damage.

RESOURCES

Load testing glossary

Plain-language vocabulary for authorized capacity testing — the metrics, test types, Layer 4 / Layer 7 terms, and compliance language teams use when planning safe stress runs on infrastructure they own.

Visual glossary of load testing terms: latency, throughput, error rate, concurrent users, soak test, abort criteria — Load testing glossary — Slammed.st resources

Why this glossary matters

Search results for “IP stresser,” “RPS,” and “load test” mix engineering vocabulary with abuse slang. This glossary keeps the authorized capacity-testing meaning of each term so product, SRE, and security stakeholders share one language before a run starts.

When teams disagree on what “concurrent connections” or “stress test” means, they misread charts, set weak abort rules, or buy the wrong plan. Clear definitions reduce false confidence and keep tests inside written permission windows.

Use these entries alongside authorized load testing, product features, and Acceptable Use. Nothing here is an attack playbook — only vocabulary for measuring systems you control.

Core metrics

Latency: Elapsed time from when a client starts a request until a useful response is received. In authorized load tests, latency is the first signal that queues, TLS handshakes, or downstream services are saturating. Always pair it with percentiles rather than a single average that hides long tails.
Throughput: Successful work completed per unit of time. For HTTP(S) that is often requests per second; for Layer 4 it may be packets, connections, or bits per second. Throughput without a latency and error budget is incomplete — high Gbps with collapsing success rate is not a pass.
RPS (requests per second): How many application-level requests the target completes each second under the current scenario. RPS answers “how much work finished,” not “how many workers we opened.” Short, cacheable GETs can show high RPS while a heavy POST path plateaus much earlier on the same host.
Concurrent connections: How many simultaneous sessions or workers the test holds open against the target. Concurrents pressure connection tables, thread pools, and keep-alive behavior even when each session is idle or slow. They are not interchangeable with RPS — see the misconceptions section and Concurrent connections vs RPS.
Error rate: Share of attempts that fail: timeouts, connection resets, HTTP 5xx, TLS failures, or refused connections. Define which codes count as errors before you ramp, and set abort thresholds so a failing run stops instead of hammering a broken path.
Bandwidth / Gbps: Volume of network traffic per second, usually expressed in gigabits. Bandwidth matters most for Layer 4 and large-payload Layer 7 scenarios. Peak platform capacity on Slammed.st reaches up to 4.3 Tb/s across regions — your plan and target pipe set the practical ceiling for any single authorized run.
p95 / p99 latency: The latency value below which 95% or 99% of samples fall. Averages can look healthy while the worst few percent of users time out. Capacity cliffs often appear first in p95/p99 before mean latency moves much — treat percentiles as release gates, not optional charts.

Test types

Smoke test: A short, low-intensity run that proves the target, credentials, methods, and dashboards work. Smoke before every campaign window so you do not discover a misconfigured host mid-spike.
Spike test: A sudden jump in load to see how autoscaling, caches, and rate limits react. Spikes are useful for launch and flash-sale hypotheses; keep abort criteria tight because recovery behavior can be worse than the spike itself.
Soak test: A longer steady run looking for memory leaks, queue growth, connection pool exhaustion, and slow degradation that a ten-minute spike will miss. Soak duration should match how long your real peak is expected to last.
Stress test: In authorized practice, a controlled push past expected peak to find the first hard cliff — not an unbounded attack. You raise load in steps, watch live metrics, and stop when abort rules fire or the capacity story is clear.
Capacity test: A structured search for the highest stable load your system can sustain within latency and error budgets. The output is usually a plateau (“stable at N concurrents / M RPS”) plus the next step that failed — useful for capacity planning and plan sizing on pricing.
Breakpoint test: Deliberately finding the first failure mode: when p99 breaches SLO, when 5xx explode, or when the NIC or connection table saturates. Document the breakpoint; do not keep driving past it without a cool-down and a written reason.

Layer 4 vs Layer 7 terms

Layer 4 (L4): Network and transport path testing — TCP/UDP connection rates, packet rates, bandwidth, firewalls, load balancers, and scrubbing. Failures often look like drops, resets, or saturated interfaces rather than HTTP status codes. Choose L4 when the hypothesis is the pipe or perimeter.
Layer 7 (L7): Application-path testing — HTTP/HTTPS, APIs, TLS, caches, gateways, app servers, and databases. You watch RPS, p95/p99, 5xx rates, cache hit ratio, and WAF challenge rates on legitimate paths you own. Choose L7 when the hypothesis is the product journey.
Method profile: The configured traffic shape for a run (protocol family, concurrency, duration, regions). Profiles should match the hypothesis: L4 for network capacity, L7 for checkout or API paths — not “maximum everything” by default.
Edge vs origin: Edge is CDN/WAF; origin is your app or API behind it. High cache-hit traffic can hide a fragile origin. Separate scenarios so you know which layer cliffed. Deeper comparison: L4 vs L7: when to use.

Guides: Layer 4 stress testing · Layer 7 stress testing.

Authorized testing & compliance terms

Verified target: A host, IP, or hostname you own or have written permission to test, confirmed through the platform’s verification flow. Slammed.st requires verified targets — anonymous “any IP” testing is not supported.
AUP (Acceptable Use Policy): The rules that define allowed and prohibited use of the service. Read the full text on Legal & Acceptable Use before any run; violations can mean account suspension and legal exposure for the operator of the test.
Abort criteria: Pre-agreed stop rules — latency ceilings, error %, CPU/NIC thresholds, or business signals — so a test ends when the system is harmed or the question is answered. Abort criteria belong in the run plan, not improvised mid-incident.
Cool-down: A quiet period after load stops so queues drain and sticky failures become visible. Skipping cool-down can hide whether the system recovers cleanly or stays degraded.
Ownership proof: Evidence that you may test the target: DNS records, control-panel confirmation, signed letters, or ticketed change windows. Keep proof with the change record; “we thought it was ours” is not authorization.
IP stresser (authorized definition): A platform that generates controlled Layer 4 and/or Layer 7 traffic against infrastructure the customer is authorized to test, with accounts, plan limits, and acceptable-use enforcement. Contrast with booters that sell anonymous disruption — see Is using an IP stresser legal?.

Platform / Slammed.st terms

Hub: The Slammed.st control panel where you configure methods, regions, concurrents, duration, and stop controls for authorized runs. Live metrics and post-run reports surface here.
Concurrents (plan limit): The maximum simultaneous workers your plan allows. Plan concurrents cap how hard you can push in one scenario; they are not a promise of target RPS. Compare limits on pricing.
Regions: Geographic sources of generated traffic. Multi-region runs exercise anycast, DNS, and edge behavior differently than a single lab box. Slammed.st offers capacity across dozens of regions for authorized tests.
Free Trial: A no-card starting plan to verify targets, learn the Hub, and run limited authorized scenarios. Upgrade when you need longer windows, higher concurrents, or API automation — register.

Feature depth: Features. Product positioning stays authorized capacity testing only.

Common misconceptions

Booter vs stresser: In street slang the words blur; in engineering they should not. A booter sells anonymous attacks on third parties. An authorized stresser (or load-testing panel) requires verified targets, identifiable accounts, and an AUP. If a panel advertises “hit any IP,” it is not a capacity tool you should buy for production rehearsals.
RPS vs concurrents: Opening more workers does not guarantee more completed requests. Slow backends, rate limits, and connection limits can raise concurrents while RPS stalls or falls. Always report both, plus latency and errors. Deep dive: Concurrent connections vs RPS.
“More Gbps” equals “app is fine”: Layer 4 bandwidth success does not prove checkout or API health. Pair L4 pipe tests with L7 journey tests when the risk is user-facing latency or error budgets.
Average latency is enough: Means hide tail risk. Release decisions should use p95/p99 (and error rate) so a small fraction of failing users is not averaged away.

Term pages

Standalone glossary entries:

Calculator: Concurrent connections vs RPS.

Related guides

Continue with these internal resources:

When vocabulary is aligned, book a verified-target window, set abort criteria, and measure — do not improvise on production without a plan.

FAQ

What is RPS in load testing?: RPS means requests per second completed against the target. It measures throughput for HTTP/API tests and should be read with latency and error rate.
What is the difference between concurrent connections and RPS?: Concurrent connections are simultaneous open sessions or workers. RPS is completed requests per second. High concurrency with low RPS often means the system is waiting or saturating.
What is an abort criterion?: A pre-agreed stop rule — such as p95 latency, error rate, CPU, or NIC saturation — that ends a stress test before uncontrolled damage.

Resources hub · Glossary · Tools matrix · RPS calculator · Useful network tools · Blog · Pricing · Start free