Service load testing

[danstiner]

Before launch we should have a target maximum number of simultaneous streamers/viewers and both be confident we can reach those numbers and have an idea of how many servers it will take to get there.

We can do some napkin math for an idea but Hayden and I chatted and agree the ideal approach would be to simulate load and see how the service behaves.

Steps

• [ ] Define a target number of simultaneous streamers and viewers we want to support at launch
• [ ] Find or create a tool that can spin up a number of ftl client streams to simulate streamers
  • https://github.com/Glimesh/ftl-sdk/tree/master/ftl_app may be relevant, existing tools probably will not exist for a basically custom protocol
  • Would it be crazy to use this as a chance to experiment with Rust since it'll probably be a separate codebase from this plugin?
• [ ] Find or create a tool that can spin up a number of WebRTC clients to simulate viewers
  • There are only a few free tools that even purport to do this, mainly https://github.com/webrtc/KITE and https://github.com/RestComm/webrtc-test
  • Article that covers using KITE: https://webrtchacks.com/sfu-load-testing/
  • I think it's interesting how webrtc-test's approach is basically to spin up instances of Chrome, maybe we could just use puppeteer to open a bunch of pages to the actual Glimesh stream pages for very realistic load?
• [ ] Run the load tests
  • We'll need multiple large, high-bandwidth servers to run the tests from

[haydenmc]

I think @clone1018 has some rough numbers for the first bullet.