0

I ran a basic stress test and am having trouble interpreting the results.

Setup

  • Super simple node.js API (returns a string for a GET request) deployed on heroku's free tier

  • Increased RPS until I started to see a lag in average response time (unfortunately the tool I was using didn't allow a p90, etc, just average)

  • Datadog integration for monitoring

While I did hit a threshold (2.5k rps) I started to see a slowdown, I didn't see anything in DataDog to indicate stress - RAM, CPU.

If it's not CPU or RAM, what is likely causing the bottleneck here? How can I tell whether vertical or horizontal scaling would be likely to help?

stk1234
  • 99
  • 2

1 Answers1

0

There can be many bottlenecks. Let me list few:

  • config limitation of application (for example you have maximum of 100 threads defined)
  • context switching (between app and kernel)
  • loaded only single core (you have 6 cores and load only one and total load is ~16%)
  • disk bandwidth
  • your test platform (where you run requests) reach limit of tests per second.
  • your provider limit you
Romeo Ninov
  • 5,263
  • 4
  • 20
  • 26