4 Replies Latest reply: Oct 16, 2012 5:29 AM by Steven Bain RSS

    Server Load Testing?!

    Steven Bain

      Hiya Guys,

       

      I am a bit lost as to the question here!  :-)  I have been asked to:

       

      "Look into planning a load test on the QV server for a max of 10 users."

       

      I have briefly investigated this webpage:  http://qliktips.blogspot.dk/2011/04/load-testing-qlikview-with-jmeter.html

      And this download package:  http://community.qlik.com/docs/DOC-2705

       

      However, it all seems rather in-depth, complicated, and comprehensive... and I am not only unsure as to how to use these packages, but also if the functionality they provide will lead to the answer(s) I seek.

       

      Does anyone have any experience in conducting this kind of test?... or what measures I should be recording to give as feedback?

       

      How to I perform a load test for a fictional number of concurrent Users?!

       

      Our QV implementation is not yet classed as "Live" - and so we probably have around 3/4 concurrent Users playing with Dashboards at any one time.  I assume I need to give feedback on how the Server will perform once there are up to 10 concurrent Users?

       

      If anyone has any information or ideas it would be greatly appreciated!

       

      Cheers,

       

      Steve.

        • Re: Server Load Testing?!
          Daniel Rozental

          Steve, yes, I believe the TestGeneratorScript is the right tool for what you're trying to do. You should spend some time getting familiar with it and building a script that's complex enough to reflect how you believe the application will be used.

           

          What you should be meassuring is CPU and memory consumption (by using windows performance monitor) and response times that will be logged by the tool.

           

          You can test 10 sessions running at the same time by configuring that in the tool. I believe that will be fairly similar as 10 different users, but if you actually want to test 10 different users then just run the script from 10 different PCs at the same time logged with 10 different users.

          • Re: Server Load Testing?!

            Hi Steve

             

            Decent link here including a link to the load simulator dashboard which will give you a very rough approximation as to CPU\ memory usage : http://community.qlik.com/message/83460#83460

             

            I went through a similar process as yourself - discounted jmeter due to the time it would take to learn and configure, discounted engaging with a reseller to implement HP LoadRunner for us to simulate 500 users due to price. If our dashboards were accessible to the internet I would have investigated a few SaaS load testing offerings out there but they're not, so I didn't

             

            For a down and dirty quick approximation try the load simulator dashboard. I've tried it on v10 SR5 and works ok.

             

            If you find anything cheap and quick let me know

             

            Thanks

              • Re: Server Load Testing?!
                Steven Bain

                Thank you Daniel Rozental and fergusona!!!

                 

                That is very helpful advice - I will look into your suggestions and have a play... I will be back in touch to share my findings  :-)

                 

                Cheers,

                 

                Steve.

                  • Re: Server Load Testing?!
                    Steven Bain

                    Hiya Guys,

                     

                    Thanks for the advice - due to time my actual solution was fairly crude (similar to Daniel Rozental's suggestion).  :-)

                     

                    As I was 'only' load-testing for 10 Users - I just got 10 colleagues to log-on to the Dashboard and click for all their worth!  I set Windows Performance Monitor away in the background (http://www.windowsnetworking.com/articles_tutorials/windows_2003_performance_monitor.html), and recorded the results.

                     

                    This very closely (if not exactly) mimicked a worst-case scenario - and, as it happens, it looks as though the server will be able to cope  :-)

                     

                    I also had a play around with fugusona's suggestion - and, although I'm not sure I was using it to its full potential - it also suggested that the Server would always be within its "Effective RAM" threshold... so I'm fairly confident we should be OK.

                     

                    Obviously if I was going to be load-testing more than 10 Users I would look more closely into the TestGeneratorScript solution.

                     

                    Cheers,

                     

                    Steve.