Hi, Thanks for your reply.
We do not want to model the Zipf-like distribution, but use it as an example of request patterns to test the Web server. We have designed some power management mechanism (i.e., dynamic voltage and frequency scaling approach) for used in Web server. We want to measure the performance of Web server (e.g., power consumption and response time) under several Web workloads, for example, Zipf-like distribution. In addition, we may generate other request patterns (not Zipf-like) and use them to test the server. So we want to control the requests issued by JMeter using a trace file. Yes, we can use the second solution and have already used it. We put all URLs into a file where the number of requests for a particular URL is based on Zipf-like distribution. Then, we use Constant Throughput Timer to control the number of requests sent per minute. However, it may be difficult to create a JMeter test plan if we want to use other request patterns that do not follow any distribution and are based on some user scenarios. That is the number of requests per second is particular, neither constant nor random. So it will be suitable for our case if JMeter can read a trace file to generate requests. JMeter is a powerful Web load testing tool but may not be suitable in our case. Please kindly let us know if other Web benchmark tools can read trace files to generate requests. Thanks for any replies. -- Best Regards, Zheng Jian-Ming a.k.a. zjm On Sun, Sep 25, 2011 at 6:28 PM, Oliver Lloyd <oliver_ll...@hotmail.com> wrote: > I see. So you want to do this based on a desire to model the zipf > distribution for page views? Here's the thing, your methodology is a little > bit masochistic - you can achieve your aims using far, far simpler means > than what you propose. > > Imagine: you would like to model a distribution of load over various pages > in order to follow zipf principles. That's great, good idea. To do this you > do not need timestamps and individual trace files per load generator - you > just need one of two things: > > 1. A series of throughput targets for each page or page type. > 2. A weighted file with lots of URLs in, ie. with some that appear more than > once. > > For 1. you would have something like this: > > Homepage - 30 tps > About Us - 3 tps > View Post - 20 tps > Submit Post - 2 tps > Register - 0.2 tps > > See, how this can be got using zipf and then used to plug into a JMeter test > plan? You'd use the Constant Throughput Timer along with Random Timers to > design a realistic model of expected load. > > For 2, if you only have static pages and no state to speak of, then you can > just lump all the URLs in one file and use the zipf formula to weight > things. You then have one sampler that reads from the same file and the > weighting along with a Random Timer gives you what you need. > > I'm sure others can provide a complicated solution to have JM read trace > files but remember, this is a tool to run load tests, if your goal is to > load test your website then why re-invent the wheel? > > > > ----- > http://www.http503.com/ > -- > View this message in context: > http://jmeter.512774.n5.nabble.com/Generate-requests-at-specified-time-tp4836359p4838174.html > Sent from the JMeter - User mailing list archive at Nabble.com. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: jmeter-user-unsubscr...@jakarta.apache.org > For additional commands, e-mail: jmeter-user-h...@jakarta.apache.org > > --------------------------------------------------------------------- To unsubscribe, e-mail: jmeter-user-unsubscr...@jakarta.apache.org For additional commands, e-mail: jmeter-user-h...@jakarta.apache.org
