Skip to content

Latest commit

 

History

History
59 lines (39 loc) · 2.53 KB

load-testing-with-ab.md

File metadata and controls

59 lines (39 loc) · 2.53 KB
Raw

Load testing with ab

Beware: for real load tests you should use an external service which generates actual load from different IP addresses etc. This testing gives you some idea of the load handling capabilities of your web app.

For this quick test, we'll be using a tool called ab.

ab is a small tool by Apache for generating concurrent requests to a HTTP url. Documentation: http://httpd.apache.org/docs/2.4/programs/ab.html

Install on OS X: brew install ab

Good response times

Everything depends on how fast / performant you want your API to be, but I would roughly say that a fast, high available API should fill these requirements:

  • Median response time should be <50ms

  • 95th percentile response time should be <100ms (95% requests handled under this repsonse time)

  • Handle at least ~1000 requests per minute (17 req/s).

    If your API can't handle that much, you should do caching or other measures to enable proper performance.

    1000 req/min is the minumum, but you should aim for handling at least ~5000 request per minute.

Running a test

Let's execute a basic test which sends a lot of GET requests to a single HTTP url.

ab -vv -k -n 1000 -c 50 http://app.herokuapp.com/api/v1/some/resource

  • -vv Verbose
  • -k Enable the HTTP KeepAlive feature, i.e., perform multiple requests within one HTTP session. Default is no KeepAlive. Enable it because browsers use it too
  • -c 50 Number of multiple requests to perform at a time. Default is one request at a time.
  • -n 1000 Number of requests to perform for the benchmarking session. The default is to just perform a single request which usually leads to non-representative benchmarking results.

Then you get this: Requests per second: 110.63 [#/sec] (mean) and it will report what the respond times were, you should aim them to be under 100ms

Is my app able to handle 20 000 monthly users?

You need to have some kind of estimate what the X amount of users in month means as requests per minute or requests per second.

So for example, if you have 20k users per month. You should figure out the max peak of requests at any given minute. If you have Google Analytics, check the hourly page views for the last month / year. Find the busiest hour from the analytics history. Come up with a pessimistic estimate what could be the largest amount of requests per minute. E.g. if in 1 hour you have had 5k page views, it would mean 83 req/min - but better make sure you can handle 500 req/min to handle peaks. Everything depends how large of a traffic burst peaks you have.