Hi Jian, I'm not sure that I follow your question. Inside a single thread in jmeter, the http samplers are executed serially. That is, the first sampler finishes completely, before the second sampler request is made. So the second sampler would not start until after the response from the web server had been completely received, i.e., after the total response time. There woud be no interactions between these transactions at all as they would never be concurrently executed. If you want additional time between the end of one transaction and the start of the next (to simulate user "think time", then you can use jmeter timers to accomplish this task. When you use multiple threads, either within one thread group or by using multiple thread groups, then requests will be made in parallel. Jmeter tracks these requests separately, so the log would contain latency and total response time for each request. Although there is probably some slight inaccuracy in jmeter's recording of these times when a large number of transactions are occurring at the same time, I think the errors are very small. For reasonable transaction rates I don't think you need to worry about contention/interactions on your PC garbling up the results. Of course your server will need to be able to handle the simultaneous requests, and there may be resource contention on that end that may cause a transaction to get slowed down as a result of other transactions being handled in the same time window. However, this is exactly what you are typically trying to measure when using jmeter... how much degradation in response time do I get when I start sending a lot of requests at my web server. There is a third area that can cause interactions between concurrent transactions. Your connection to your internet service provider, or your server's connection to the internet will have bandwidth limitations that define how much data can be sent and received over the connection at the same time. You should look at the throughput figures (KB/sec) in the aggregate report to make sure that this figure is a lot lower than your bandwidth to the ISP. If not, then your results could get skewed a lot by network contention. I don't know if this answered your question or not. If not restate your question. Alan ________________________________
From: Jian Tong [mailto:[EMAIL PROTECTED] Sent: Tuesday, December 19, 2006 8:56 AM To: JMeter Users List Subject: RE: Any way to get more response time breakdowns using Jmeter? Hi Alan, I got another question with regards to elapsed time and latency. Suppose I have two requests inside one thread, there would be two respective elpase time and latency for each request in the summary. I am wondering how the two requests will affect each other on their elapsed time and latency ? How reliable the results would be ? The senario I am dealing with is: users send requests to a game site and play the game, then send their scores to the server in the end. Thank you Jian >>> "Christensen, Alan" <[EMAIL PROTECTED]> 12/18/06 3:39 PM >>> Hi Jian- Yes it would be very nice if they included latency in the summary. I only just found out recently that "time to first byte" was collected by jmeter because I didn't realize that they were using the word "latency" to refer to that. Most test systems use the term "time to first byte" to refer to this measurement. I've also seen the term "server busy" used to refer to this measurement, as it has a closer correspondence to the time when the server completes processing the request. It would be very nice for someone to request via a bugzilla enhancement request your suggestion to include latency in the summary report. Would you do this? Cheers, Alan ________________________________ From: Jian Tong [mailto:[EMAIL PROTECTED] Sent: Monday, December 18, 2006 10:54 AM To: JMeter Users List Subject: RE: Any way to get more response time breakdowns using Jmeter? Alan, thank you so much for your detailed explanation. As latency is even a better choice than elapased, shouldn't it be included in the 'summary report' and 'aggregate report' ? In this case, we can use both times for the measurement of the systems' performance. Thanks Jian >>> "Christensen, Alan" <[EMAIL PROTECTED]> 12/17/06 7:00 PM >>> Latency is the time at which the first byte of the html page gets received by jmeter. After that point in time several TCP/IP segments may be required to actually download the entire page. These segments may require acknowledgements which would introduce additional network delays before the page is fully received. In addition, the bandwidth of the "last mile" will typically limit the speed at which the rest of the page gets delivered. The latency time is typically a better measure of system response time than the total response time. This is especially true of large pages where the network component of the response time can easily dominate the user response time. One thing I noted is that the latency time, and many other parameters, do not show up in the log unless the xml format is selected. This makes it much more difficult to get the log into an excel spreadsheet than is the case with a simple .csv log. I'm assuming that this limitation is a bug in jmeter. Does anyone know if this is the case, or if there are any workarounds? -----Original Message----- From: Jian Tong [mailto:[EMAIL PROTECTED] Sent: Sunday, December 17, 2006 1:27 PM To: [EMAIL PROTECTED]; jmeter-user@jakarta.apache.org Subject: Re: Any way to get more response time breakdowns using Jmeter? I am still not clear what latency time is. Anybody can give a definition of latency time in JMeter ? The sample xml log in User Manual shows Latency time is much less than Elapsed time. How is Latency related to Elapased time ? Thanks Jian ------------------------------------------ Jian Tong - QA Production and Operations, CBC.ca Extension: 3357 Location: 2C400-N ------------------------------------------- >>> sebb <[EMAIL PROTECTED]> 12/15/06 9:59 AM >>> On 14/12/06, Christensen, Alan <[EMAIL PROTECTED]> wrote: > Most of the commercial testing tools provide a lot more information with > respect to response time. For example, most of them report the response > time to the first byte as well as the response time to the last byte. > Breakdowns of DNS time and content downloads are also typically > included. Does Jmeter have any of these capabilities, and if so, how > would I turn them on? See: http://jakarta.apache.org/jmeter/usermanual/listeners.html some listeners support latency (time to first response). Other timings are not currently collected - please feel free to create a Bugzilla enhancement request with full details of what you are seeking. Patches also welcome. S. --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
