Distributed load testing with Gatling, is it possible???

I have been doing some load test with gatling 1.5.1 and so far i am pretty happy with the framework.
I was wondering though if anybody had ever done distributed load testing with this framework.
I know gatling uses akka actors, maybe someone has already play with it and make this work.


We run distributed load tests in Gilt; we built a (slightly hacky) set of scripts run by Jenkins that create EC2 instances, upload executable jars to them, run the load test without generating reports on each one (-nr, I think), and harvest the simulation logs afterwards. Once you have a directory of simulation logs you can create a report and test assertions against the aggregate of them by just using -ro (and -s if you have assertions to test).

Thanks for the response Andrew.
I am not sure I follow completely, do you have some examples or tutorial of what your set up is doing?
I see gilt allows you to connect to another host within gatling, and i can easily create some virtual machine. I am not sure how the scenario will actually run on the machines?

On Thursday, June 20, 2013 2:16:20 PM UTC-7, Assane Diop wrote:

Thank you Andrew.
May I suggest that this would be a nice topic for a post on Gilt’s blog (excellent material there, BTW, you guys rock!)?

Could Gatling run distributed load testing now?

在 2013年6月21日星期五 UTC+8下午1:30:18,Stéphane Landelle写道:

Distributed load test is implemented albeit manually. You can run the simulations from various machines, collect the log and create a report. Check out http://gatling.io/docs/2.1.7/cookbook/scaling_out.html

What Andrew has done is create scripts to do all these manual tasks. We also did something like this with AWS EC2 instances using fabric.

Gatling Team,

are there any plans to provide any ‘cluster’ or ‘distributed’ mode? Do you have any road map for this year?


W dniu wtorek, 28 lipca 2015 20:09:19 UTC+2 użytkownik Abhinav Gogna napisał:

Absolutely, our upcoming commercial product, FrontLine, will support clustering.

Hi Assane,

Until the commercial product arrives, I’d probably use a lightweight distribution tool[1] and aggregate hosts in InfluxDB. Unless, the Gatling reports were actually required, I’d probably turn them off and disable the parser.

I can write some notes on this if required?

Can I ask what your expected concurrency peak is?

[1] https://github.com/bbc/notes-on-perf-testing#distribution


One of the issues with InfluxDB and other standard time series databases is that if you care about percentiles/distributions (which you should, those are the only metrics that make sense, on contrary to average and standard deviation), you end up averaging them (over time, over cluster nodes).

This is of course completely wrong (e.g.: the 100th percentile is the maximum, and the global maximum is absolutely not the average of the local maximums) and artificially smoothen your results.

This is one of the issues we’re fixing with FrontLine.

That is a very good point.