What is a load generator, or load engine?
A load generator (Loadster calls it a load engine) is the machine, or fleet of machines, that runs your virtual users (bots) during a load test and fires their traffic at the system under test. In Loadster, each group of bots runs on either a Loadster Cloud region, which is managed infrastructure launched on demand, or a self-hosted Loadster Engine that you run on your own network.
Why Load Generation Is Its Own Problem
The part newcomers to load testing often overlook is that simulating large numbers of users takes real CPU, memory, and bandwidth. A big test can require thousands of cloud instances, and real browser load testing (Browser Bots or Playwright) is especially resource intensive because each one drives a real web browser, which consumes far more resources than a lightweight HTTP client (Protocol Bots).
Regardless of the load testing tool you use, undersizing a load generator doesn’t just slow the test down, it can invalidate the results!
An overloaded load engine reports inflated response times that look like server slowness, so you might end up blaming your site for a bottleneck that lives on the testing side. Loadster’s cloud sizing exists to ensure ample resources are available to run the bots themselves; see the cloud capacity FAQ for how Loadster sizes clusters in each region for your load tests.
Cloud Regions vs Self-Hosted Engines
For testing internet-facing sites and web applications at scale, cloud regions are usually the best choice. Loadster launches instances on demand in your choice of 32 regions across AWS and GCP, typically in about 1-3 minutes. Picking regions close to your real users makes response times more representative of reality, and spreading a large test across several regions surfaces geographic latency differences.
For testing applications on a private network behind a firewall, where the cloud regions can’t reach, you can run the self-hosted Loadster Engine instead. Self-hosted engines are packaged as a Docker container that runs on your own hardware, makes an outbound connection to Loadster’s cloud platform, and generates load from inside your network at no additional charge. There’s a walkthrough in our FAQ on load testing a site on a private network.
How Many Load Engines Do You Need?
With cloud regions, Loadster handles the sizing automatically: the platform launches however many instances your bot counts require, and warns you in the scenario editor if you’ve allocated more bots to a region than it can likely provide.
With self-hosted engines, sizing is for you to figure out. We recommend at least 2 CPU cores (preferably more) and 2 GB of free RAM per engine at a minimum, with a fast network connection, and keeping an eye on the engine’s CPU utilization during tests. For real browser testing you’ll need much more resources.
If one machine isn’t enough, you can run multiple engines and split your bot groups across them.
One more thing to keep in mind: if your site sits behind a WAF or rate limiting gateway or proxy, you’ll usually want to allow the test traffic through before running anything big. Check out the FAQ on whitelisting load test IP addresses for how to do this.