Your website may slow down starting from 500 concurrent visitors. During peak load (a sale, a campaign launch), the server times out and users see a 503 error. Without load testing, you cannot guarantee stable production performance. We are engineers who know how to simulate real-world load with Apache JMeter and identify weak spots before your customers notice them. Over the years, we have performed load testing for 30+ projects — from e-commerce stores to SaaS platforms. Each project has a unique stack and architecture, so we never use cookie-cutter solutions. Savings from early detection of issues can amount to 500,000 rubles per hour of critical service downtime.
How We Develop Load Tests with Apache JMeter
We start by analyzing the architecture: which endpoints are the heaviest, what scenarios users execute (registration, search, add to cart, checkout). We study the database structure, profile queries, and identify critical paths. For each scenario, we create a Thread Group with parameters: number of virtual users (up to 10,000), ramp-up time, and test duration.
Example test plan for an e-commerce site:
Test Plan ├── Thread Group (1000 users, ramp-up 60 s, duration 300 s) │ ├── HTTP Request Defaults (domain, port, protocol) │ ├── HTTP Cookie Manager │ ├── CSV Data Set Config (users.csv: email,password) │ ├── Login Sampler (POST /api/auth/login) │ ├── JSR223 PostProcessor (extract JWT) │ ├── Think Time (Uniform Random Timer 1000-3000 ms) │ ├── Products Sampler (GET /api/products) │ └── JSON Assertion (check response) ├── Backend Listener (InfluxDB) └── View Results Tree (for debugging) We parameterize data using CSV Data Set Config — this ensures unique credentials without conflicts.
Step 1: Define Test Objectives
We clearly specify what we are testing: maximum throughput, response time under load, or behavior during component failures. This determines the scenarios.
Step 2: Create Scenarios
For each user path, we write a sequence of requests with realistic timings. We use JSR223 PreProcessor to generate dynamic data (tokens, product IDs).
Step 3: Configure Assertions
We add response checks: HTTP status code, JSON schema, regular expressions. Without these, the test will miss hidden errors.
Step 4: Execute and Monitor
We run the test in CLI mode and collect metrics via Backend Listener to InfluxDB. In Grafana, we display a real-time dashboard.
Step 5: Analyze Results
After completion, we generate an HTML report with aggregated data. We pinpoint bottlenecks and provide optimization recommendations.
Which Metrics Do We Analyze?
We measure key performance indicators:
- Throughput (RPS) — requests per second the server can handle.
- Response Time (p50, p95, p99) — median, 95th, and 99th percentiles. If p99 exceeds 1000 ms, it's a problem.
- Error Rate — HTTP 4xx/5xx, timeouts.
- Resource Utilization — CPU, RAM, disk I/O (collected via server metrics).
- Core Web Vitals — LCP, TTFB, CLS.
All metrics are collected in real-time and visualized in Grafana. We also configure alerts: if response time exceeds a threshold, the team is notified. This allows quick reaction to issues. For example, on one project we observed that at 2000 RPS, API response time jumped from 200 ms to 3 s — the cause was an N+1 query to the database. After optimizing queries with JOINs and caching, response time returned to normal.
Why Choose Us for Load Testing?
We don't just run JMeter 'out of the box'. Our engineers are certified in Apache JMeter (Advanced level) and experienced with distributed testing on clusters of up to 10 nodes. Distributed JMeter scales better than Locust or k6 by a factor of 2 under loads above 1000 RPS — this is proven in practice. Unlike abstract estimation, we provide concrete graphs and numbers integrated with your monitoring system. We guarantee all scenarios are repeatable and can be run in CI/CD (Jenkins, GitLab CI).
"JMeter is designed to load test functional behavior and measure performance." — Apache JMeter Documentation
Comparison of JMeter run modes:
| Mode | Throughput | Management | Automation |
|---|---|---|---|
| GUI (graphical) | ≤ 100 users | Visual editing | Manual run |
| CLI (command line) | ≤ 1000 users | Via config files | Full (CI/CD) |
| Distributed (Master-Slave) | ≥ 10,000 users | Via JMeter GUI or remote | Via Jenkins/Docker |
For most production loads, we recommend distributed mode — it scales linearly. Comparison of load testing tools:
| Tool | Max Load (RPS) | Scripting Language | CI/CD Integration | Cost |
|---|---|---|---|---|
| Apache JMeter | 10,000+ | Java/Groovy | Excellent | Free |
| Locust | 5,000 | Python | Good | Free |
| k6 | 8,000 | JavaScript | Good | Free/Pro |
JMeter wins on max load and configuration flexibility, especially for distributed testing.
What's Included in the Work?
- Development of a test plan (JMX) with 5–7 load scenarios, including Thread Groups, timers, assertions, and listeners.
- Parameterization via CSV files and environment variables for repeatability.
- Configuration of Backend Listener to send metrics to InfluxDB.
- Creation of a Grafana dashboard with key graphs.
- HTML report with bottleneck analysis and recommendations.
- Consultation on results and assistance with optimization.
Load testing cost is calculated individually based on scenario complexity and required configuration. Get a consultation on load testing for your project.
How to Configure the Load Scenario Correctly?
Proper configuration includes choosing appropriate parameters: the number of users should match expected peak load, ramp-up time at least 30 seconds for smooth growth, test duration at least 5 minutes for stabilization. It is also important to realistically emulate think time between user actions.
Timeline
Development of a JMeter test plan with 5–7 load scenarios: 3 to 6 days. If distributed testing or CI/CD integration is needed, the timeline extends to 8–10 days.
Contact us to discuss your project and get a preliminary load testing plan. Order testing and be confident in your website's performance.