search

mcollina / autocannon

fast HTTP/1.1 benchmarking tool written in Node.js
MIT License
7.83k stars 325 forks source link

Configuring for high concurrency #113

Closed behrad closed 7 years ago

behrad commented 7 years ago

How we can simulate a real world load? Where concurrent users (=connections) are high, each make 3-4 requests, close. another user...

Does this really mean 10k users each issuing 2 requests? autocannon -c 10000 -p 1 -d 180 -t 15 --maxConnectionRequests 2

mcollina commented 7 years ago

This tool is built to generate load on your application. By using HTTP pipelining and reusing the same sockets, you can generate more load than without. It is built to prove that your application can hold a certain level of throughput.

If you want to very the TCP stacks, OSS network configuration, and verify a more real-world scenario, I recommend to use bees-with-machineguns, or other tools. They generate less load per host, but they reproduce a scenario closer to real-word by using multiple VMs on the cloud.