How to Conduct a 50,000-User Concurrent Load Test with JMeter and BlazeMeter

This article provides a comprehensive tutorial on performing a large‑scale load test targeting 50,000 concurrent users using JMeter and BlazeMeter. It starts by emphasizing the need for the latest JMeter version and optional plugins, then describes three ways to obtain test scripts: BlazeMeter Chrome extension, JMeter HTTP(S) Test Script Recorder, or manual creation.

Step 1: Write Your Script

Ensure you have the latest JMeter release from the Apache site and install helpful plugins. When creating scripts, parameterize user‑specific data (e.g., usernames, passwords) using CSV files, regular expressions, JSONPath, or XPath extractors, and use configuration elements to simplify environment switches.

Step 2: Local Testing with JMeter

Run a single‑thread, single‑iteration test using the View Results Tree listener to debug samples and logs. Verify all scenarios, then scale up to 10‑20 threads for a longer run, checking response times, errors, and throughput.

Step 3: BlazeMeter Sandbox Test

Configure a sandbox test with 300 users, one console, and a 50‑minute duration. Ensure firewall CIDR lists are whitelisted, all test files (CSVs, JARs, JSON, etc.) are accessible, and no local file paths are used. Monitor logs for errors.

Step 4: Set Users per Engine

Run a test with 500 threads, 40‑minute ramp‑up, unlimited iterations, and a 50‑minute duration using one console and one engine. Observe CPU and memory usage; if usage stays below 75 % CPU and 85 % memory, increase threads, otherwise adjust down.

Step 5: Install and Test a Cluster

A cluster consists of one console and up to 14 engines. Test with increasing engine counts (up to 14) while ensuring no engine exceeds 75 % CPU or 85 % memory and the console stays below those thresholds. Record aggregate statistics to determine cluster throughput.

Step 6: Use Master/Slave Mode for Maximum Concurrency

Assuming each engine supports 500 users and each cluster can host 12 engines, achieving 50 k users requires roughly 8.3 clusters. The guide recommends using ten clusters with 10 engines each (5 k users per cluster) for simplicity and scalability. It then details how to create a Master test and nine Slave tests, rename them, switch the Advanced Test Properties from Standalone to Slave/Master, and finally run the Master‑Slave configuration to generate the desired load.

After setting up the Master‑Slave tests, you can start the Master to launch ten parallel tests, each handling 5 k users, and optionally customize scripts, CSV data, network emulators, or regions for each test.