What each metric measures, what it reveals, and how to read it, from request throughput and tail latency down to network and load generator signals. For performance engineering, platform, and QA teams.
Modern teams need more than pass/fail results. They need rich, contextualized metrics to diagnose bottlenecks and validate resilience at scale, from request-level behavior down to network and infrastructure signals.
These eight metrics cover three layers. Read them together, and you can tell a slow backend from a saturated network or an overloaded load generator.
Tracks the volume of requests sent to the system and responses returned every second during a load test.
Throughput trends reveal system capacity under increasing load. If the request curve keeps rising while the response curve plateaus or drops, it usually signals thread pool exhaustion, backend saturation, or queue overflows.
Overlay throughput with concurrent users and error rates to find the exact inflection point where performance starts degrading.
Measures latency for the slowest users, typically the 95th and 99th percentiles, rather than averages.
Percentiles expose tail latency, which is what real users experience in worst-case scenarios. Averages can mask these spikes entirely. A rising p99 with a stable mean is a classic early warning sign of a regression.
Always track p95 and p99 during ramp-up and peak phases. Tail latency is what breaks SLAs and drives churn, not averages.
Monitors the volume of errors (such as 4xx and 5xx) per second and breaks responses down by HTTP status code.
Error spikes correlate strongly with system instability. 5xx surges after peak load point to capacity or dependency failures, while growing 4xx rates may reflect routing or client issues.
Slice errors by endpoint and scenario to find the root cause quickly. A sudden 5xx spike combined with flattening throughput often means the backend is collapsing.
Tracks how many users start (arrival), finish (termination), and are active at the same time (concurrency) during the test.
These metrics give a clear picture of load patterns and system response. Diverging arrival and termination curves often indicate slow transactions, timeouts, or abandoned sessions.
Use concurrency alongside response percentiles to pinpoint capacity ceilings. Unexpected drops in termination rate usually point to blocking issues.
Measures the end-to-end duration of defined groups of requests, typically mapped to user journeys or business transactions such as checkout, search, or login.
Group metrics expose compounded latency across services and steps that single request timings miss. One slow API might not break an SLA, but several small delays add up quickly.
Use group percentiles to benchmark real business flows, so you prioritize the optimizations with the most customer impact.
Measures the time taken to establish TCP connections from the load generator to the system under test.
Rising TCP connect times typically point to network or infrastructure bottlenecks, such as load generator saturation, exhausted connection pools, or firewall limits, before application performance visibly degrades.
Watch TCP connect percentiles during ramp-up. If they rise before response times do, the issue is likely outside the application layer.
Tracks the duration of SSL/TLS handshakes while secure connections are established.
Handshake latency often becomes significant at scale, especially with CPU-heavy ciphers or certificate chain issues. Sudden increases during high-concurrency tests can add unexpected overhead.
Baseline TLS handshake times early and watch for spikes under load. They're often invisible in APM tools but visible in Gatling's client-side view.
Monitors CPU usage and memory consumption on the load generators themselves, to make sure the test infrastructure can sustain the target traffic.
If generators hit CPU or heap limits, they become the bottleneck, adding artificial latency or failing to maintain target injection rates.
Always check generator health before blaming the system under test. Sustained CPU above 80% or heap usage above 85% can skew results dramatically.
Once you track the core metrics, the next step is to connect individual test results with long-term trends and production telemetry.
Load testing isn't just about validating one release. Comparing performance across many runs reveals subtle regressions, capacity drift, and architectural bottlenecks that a single test misses.
Gatling Enterprise Edition's dashboards let you compare runs side by side and track key metrics over weeks or months, so you catch degradation before it becomes a production incident.
Performance metrics gain their full diagnostic power next to infrastructure and application telemetry. Native integrations with Datadog and Dynatrace overlay load test data with real-time observability signals.
The correlation shows whether a slowdown comes from code, network saturation, database contention, or infrastructure limits, and gives developers, SREs, and QA one shared view.
Your users expect speed and reliability, and your load testing platform has to deliver the same. Gatling Enterprise Edition is built around five pillars.
Real-time visibility with dashboards, trend comparisons, and actionable insights.
Code, low-code, or no-code: script in JavaScript, TypeScript, or Java, import Postman, or design visually.
Trigger simulations from CI/CD or the API, apply stop criteria, and gate releases with performance thresholds.
RBAC, SSO, quotas, and shared reports. Share results in Slack, Teams, or Jira.
Run tests from Gatling-managed regions, your own cloud, or on-prem.
Take all eight metrics and their pro tips to your team, or see them live in Gatling Enterprise Edition's dashboards.
All eight metrics, what each one reveals, and the pro tips to read them right.
PDF · free, no form Download the PDF downloadA guided walkthrough with a Gatling engineer, with time for your questions.
Join the next live session arrow_forward