Forum Discussion

eliao1978's avatar
eliao1978
New Contributor
13 years ago

How to interprete the page response time?

I have a quick question regarding the page response time in the test report. I understand the way how loadComplete measures page load time is different from staring at the browser and count until the web page is rendered. But I would like to get some clarifications and make sure didn't miss anything.



I ran a 30-minutes-test with 1 virtual user. As I can see in the attached test report, the average page load time is 6.7 seconds. I also tried to replicate the same scenario using FireFox browser with FireBug. I cleaned-up browser cookie/cache before making requests. I added up the elapse time of each request (request to html, js, images, css, etc.) and the average page load time was around 3 seconds. What would cause such difference?



Thanks!

7 Replies

  • irina_lukina's avatar
    irina_lukina
    Super Contributor

    Hi Eric,


    The fact is that LoadComplete doesn't use the cache at all. It always loads all the resources (images, scripts, and so on) used by the page from the server. However, when you use Firefox, the browser doesn't use the cache only for the first page. After the first page is loaded, the browser starts loading common resources from the cache, but not from the server. This makes the page be loaded faster.


    Also, browsers typically load page resources using several threads. Such parallel loading also reduces the overall page load time. However, LoadComplete loads page resources in turn (in consecutive order), and this fact can also influence the page load time shown in the attached image.


    I hope this information helps :)

  • eliao1978's avatar
    eliao1978
    New Contributor
    Hi Irina



    Thanks for your information. Is it possible to configure loadComplet requesting page resources simultaneously? If not, how should I correlate and utilize the page load time in loadComplete test report to determine whether my web application meets the performance requirement?         
  • mpunsky_1's avatar
    mpunsky_1
    New Contributor
    Hi Eric,



    What type of application are you testing against? 


    • If it is RIA, the concept of pages, and page timings gets very blurred


    You ask about load page resources in-parallel, which resources are you interested in loading at the same time?


  • eliao1978's avatar
    eliao1978
    New Contributor
    I wonder whether I can configure loadComplete to load page resources such as css, js, images simultaneously in order to emulate the real browser behavior. I was told that because loadComplete loads page resources in the consecutive, the page load time in the test report is higher. I want to know whether I can still rely on loadComplete's  page load time as the KPI or maybe I need to comprehend the number in a different way.

              
  • Anonymous's avatar
    Anonymous
    Hi Eric

    the option to issue requests in parallel to emulate an actual browser is not available in LoadComplete at this point. It is something that we are considering for implementation.

    What kind of KPIs you have for your web app performance?
  • eliao1978's avatar
    eliao1978
    New Contributor
    Hi Sergel



    Thanks for the clarification. That shall help us understand the test result better moving forward.



    The KPIs we currently focus on is the page load time. The web app is a dashboard-type of app which contains lots of client side js scripts, images, videos, and flash. The page load time from the initial test was much higher than we thought because the way loadComplete operates. That is why I asked whether loadComplete can emulate as an real browser and measure performance based on that fashion. Given the fact that loadComplete emulates requests sequentially and then aggregate those number at the end, how do people usually interpret the page load time from the report and correlate that number with web app's performance matrix?                     
  • Anonymous's avatar
    Anonymous
    Eric



    we recognize that there is room for improvement in accuracy of simulation that is desirable for high-content and JS-rich pages, and we are working on that.



    First, as we now recommend, it would make sense to zero out think time (there is an option for that) and manually set think time for page requests (which should be relatively few). This is a somewhat more accurate setup for highly parallel page loads.



    In those types of scenarios, LoadComplete will give you an upper bound of page load time while distributing load on the server over time that is, on average, representative of the user traffic captured in a scenario. That's one interpretation that is used.



    In other cases, mostly for optimization, people use the waterfall charts and then compare relative page load times after they change / reorder code to change the sequence of internal events on a page.



    Hope this is of help.



    Regards,