>I don’t quite understand your calculation. I browsed through our web site and saw that each page requires from 44 >to 78 requests Ah I didn't notice your original table separated out pages from requests. whereas I used the terms interchangeably. So for this mail a Page = the URL the browser requests. Resources = Embedded Javascript/CSS/Images that the browser automatically requests I generally tend not to test out anything related to CSS/Javascript/Image(unless Im stressing the webserver) while doing any response time testing so while a Page may include 50 resources I would only figure out values for the page. My reasons are here - http://theworkaholic.blogspot.com/2010/02/random-stuff-on-jmeter-testing.html
>If we assume that user views 2 pages a minute, than a single user would generate 500 requests Yes but then the average time per resource would be much less than 2 seconds (I just chose 2 seconds as a number, you can use the value appropriate for you). The point is that you need a minimum number of concurrent threads to generate a load and that you would observe your system with different values of concurrent threads. The initial values are just rules of thumb. >By the way, I checked and on average a page loads in around 6 seconds (fully refreshed). If you are using firebug then you should be using the network tab and looking at the times there. A javascript/DHTML heavy app may take more time to load but Jmeter cant capture this. >Sorry, lost in terminology here. How can a thread generate only 1 >connection? Each thread at least clicks on one page, which generates 50 >requests on average (see above), therefore 50 connections (in my >understanding: 1 browser request = 1 server socket = 1 connection). A browser can reuse sockets (keep alive - but we'll ignore that for now ). In any case its the number of concurrent connections that usually cause problems. So if a browser makes a request for a page and then finds out the page has 50 resources it will use anything from 2 to N concurrent connections(It used to be that the browser would only make 2 connections per domain but I think more recent browsers have increased this value , I dont know the value of N) . JMeter on the other hand does no such thing. You either need to parse the response and make the requests yourself (usually in the same thread so concurrent connections =1) or you check the download embedded resources checkbox and the same thread will download the resources but it will one connection at a time only. >Is there a Listener that gives you data for each loop? Not as far as I know . You can either prefix some index to the sample name or you can process raw data and use something like standard deviation to see how close or spread out your values are regards deepak On Thu, Sep 9, 2010 at 4:31 PM, Prostak <[email protected]> wrote: > > Thank you, Deepak, for the well thought and detailed reply. > > >So a single user would in 5 minutes be able to make (5*60)/2 = 150 > requests > > (Assuming zero think time). > > I don’t quite understand your calculation. I browsed through our web site > and saw that each page requires from 44 to 78 requests (I used a firebug > with ctr f5). So let’s say each page on average generates 50 requests > (including all images, css etc.). If we assume that user views 2 pages a > minute, than a single user would generate 500 requests in 5 minutes (not > 150 > requests). Therefore it would take only 3 users (3 threads) to generate > 1500 > requests shown in my statistic report. > > By the way, I checked and on average a page loads in around 6 seconds > (fully > refreshed). > > >In which case you'd increase the threads but put in a constant throughput > >timer so that you still see the same load. > > How do I figure out the Target Throughput for the Constant Throughput > Timer? > > >Any number is fine , its usually used to protect the server from seeing a > >burst that it might not be able to handle. lets say the calculations above > >resulted in you having to test with 50 threads, but because of timers / > >delays etc it may turn out that any point of time the server only sees 20 > >concurrent sockets. If you have no rampup then the server sees 50 > >connections immediately which might cause your tests to get errors/ lower > >values than it should > > Sorry, lost in terminology here. How can a thread generate only 1 > connection? Each thread at least clicks on one page, which generates 50 > requests on average (see above), therefore 50 connections (in my > understanding: 1 browser request = 1 server socket = 1 connection). > > >There are a lot of transient variables (network bandwidth, application > >caches, one time static loading of classes , lazy loads by DB etc) due to > >which you shouldn't rely on a single loop. Any number that gives you > >confidence that the values you are getting are accurate (the numbers > >shouldnt wildly vary for the loops). Typically I have a run to eliminate > >some of these variables (my application uses a cache so first time hits > >result in slower db calls but next time the cache should be hit) and then > >run the actual tests. > > Is there a Listener that gives you data for each loop? > -- > View this message in context: > http://jmeter.512774.n5.nabble.com/Test-plan-for-970-page-requests-every-5-min-tp2826174p2834285.html > Sent from the JMeter - User mailing list archive at Nabble.com. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > >
