Load testing the UI

Question

I have been working on a site that makes some pretty big use of AJAX and dynamic JavaScript on the front end and it's time to start stress testing. But how do you properly stress test something that requires clicking several links on the front-end? One way I was able to easily hit every page of the site quickly and repeatedly was to point a Google Mini at it. But that's not going to click links and then navigate Modal windows and things like that.

Edit - I should point out that the site is done in PHP5 and the JavaScript library used is jQuery. Not sure if this would make any difference but felt it might be useful to know.

Answer 1

There are load testing tools that can support AJAX. For example, WebLoad

http://www.radview.com/solutions/ajax-load-testing.aspx

Answer 2

What you really want is to stress test is the server's ability to handle the ajax requests. Use a load tool that looks at the requests while "recording" the test, and then tune as appropriate. I have only used the vs test edition one, so I can't point you to a low cost one.

Answer 3

JMeter is great at this. You may record your sessions and tweak them to your liking.

So-called 'ajax load testing' is a recurring subject on this site, and is often confused. So let's get it straight: There is really no difference between load testing a normal web page and load testing with ajax. It all boils down to discrete requests; they just happen to not be full page refreshes.

One thing to keep in mind is there is a distinct difference between load testing the server processing the requests (a load test) and the performance on screen of the UI components being updated (how well your javascript performs.)

Simple load test example:

1. initial page load
2. login
3. navigate?
4. 5-10 'ajax' requests (or whatever may fit your application usage pattern)
5. logout

Answer 4

I disagree with Nathan and Freddy to some degree. They are correct that "AJAX testing" is really no different in that HTTP requests are made. But it's not that simple. See my article on Ajaxian.com on Why Load Testing Ajax is Hard.

JMeter, Pylot, and The Grinder are all great tools for generating HTTP requests (I personally recommend Pylot). But at their core, they don't act as a browser and process JavaScript, meaning all they do is replay the traffic they saw at record time. If those AJAX requests were unique to that session, they may not be suitable/correct to replay in large volumes.

The fact is that as more logic is pushed down in to the browser, it becomes much more difficult (if not impossible) to properly simulate the traffic using traditional load testing tools.

In my article, I give a simple example of how difficult it becomes to test something like Google's home page when you want to query 1000's of different search terms (an important goal during load testing). To do it with JMeter/Pylot/Grinder you effectively end up re-writing parts of the AJAX code (in your case w/ jQuery) over again in the native language of the tool.

It gets even more complex if your goal is to measure the response time as perceived by the user (which is arguably the most important thing at the end of the day). For really complex applications that use Comet/"Reverse Ajax" (a technique that keeps open sockets for long periods of time), traditional load tools don't work at all.

My company, BrowserMob, provides a load testing service that uses Firefox browsers, powered by Selenium, to drive hundreds or thousands of real browsers, allowing you to measure and time the performance of visual elements as seen in the browser. We also support traditional virtual users (blind HTTP traffic) and a simulated browser (via HtmlUnit).

All that said, usually a mix of a service like BrowserMob plus traditional load testing is the right approach. That is, real browsers are great for a full-fidelity load test, but they will never be as economical as "virtual users", since they require 10-100X more RAM and CPU. See my recent blog post on whether to simulate or not to simulate virtual users.

Hope that helps!

Answer 5

You could use something like openSTA.

This allows a session with a web site to be recorded and then played back via a relatively simple script language.

You can also easily test web services and write your own scripts.

It allows you to put scripts together in a test in any way you want and configure the number of iterations, the number of users in each iteration, the ramp up time to introduce each new user and the delay between each iteration. Tests can also be scheduled in the future.

It's open source and free.

It produces a number of reports which can be saved to a spreadsheet. We then use a pivot table to easily analyse and graph the results.