Load Generator issue
we have a big problem with load generators: we have two types of remote machines (Windows 10 and Windows Server 2012 R2), and their behavior is different with different types of applications. For example, with a Web HTTP-HTML navigation, if we use the Windows Server's machines as generators we obtain an high transaction response time, instead if we use the other type of generators we can obtain better performances.
We have done many monitoring tests about the single machines and there aren't CPU, Memory or web problems. Also, we tried to restart the LR Agent process and the machine itself, but the results are the same, bad transaction response time. You can see the image below as an example of our results (an application recorded with Web HTTP-HTML protocol).
Can you give us some suggestions about it?
Thank you very much,
Too, a year or two ago, I'd determined empirically that Server 12R2 was vastly superior to Server 08 (I know this is irrelevant to your question) by using HP Diagnostics' metrics. I'd have to dig up the emails I wrote for the specifics.
I see your premise is just the opposite, however, so I'm confused. Although, now I see that you're reporting an issue with the Login transaction. I'd further granularize the login process to see if there's a step which is causing the problem on the Server LG. Otherwise, if the login is already only a single step, I'd be looking at the backend to see if there's something to be found in the application's logs. Too, the data being used to log in could be suspect. As with most things software, there are so many different components which could be the root cause and further speculation on my part is fruitless.
thank you very much for your response!
In effect, with this particular application (of which, from logs, we haven't found any problem) we obtain better performance with Windows 2010 machines instead of Windows Server as load generators.
We have also tried to granulate the login step, composed by two calls, but the results still the same: bad average transaction response time. We don't know where can be found the problem.
I ask if someone can send us some suggestions about.
Could it be a network condition problem on the Win Server? I would suggest you to monitor the response time of your login step in Chrome browser > developer tools > network tab, on both machines, to check if there is also a significant difference.
Btw, you may check the raw data in Analysis to get the consititution of transaction response time and see whether think time is included in any of the charts.
thank you for the response.
The response time of the application from the browser (F12) - Chrome, IE, Firefox -, is always low, with the times recorded with Windows 2010 machines (as you can see in the image from the first message).
Analysis doesn't impact on results, we notice the performance's difference immediately, after few seconds since the run starts from the Controller.
We are groping in the dark, we do not know where to locate the cause after all the tests carried out.
@lb1You need to look carefully to all kind of differences, like LR-runtime settings, system HW, NW, (domain) policies pushed by your IT organization, virus scanner etc. Take performance monitoring data during 'idle' and LR-run time and compare metrics.
It might even have to do with switches/routers. So switch the network cables between the two computers etc.
One of the major differences between a server and desktop based Windows installation is related to performance characteristics (E.g. foreground processing vs background processing) etc.
The suggestion to compare the F12 chrome recording is a good one.
This is a process that is hard to support via a forum there are too many variables involved.
Reward community members who take time to respond and help.
Run the script in VuGen, you will see the transaction time breakdown instantly in the output pane. Then let us know the difference of those lines.
Thanks to Tedy and JHF for the interest on my problem.
We tried to use JMeter from the load generator (Windows Server): we launched the same HTTP-HTML navigation from JMeter and average response time is very fast, as similar as the real browser navigation, so different between Load Runner results; we have also controlled Runtime Settings, browsers F12 and Analysis tool but there're no filter, in fact we have similar results between JMeter and Windows 10 load generators.
As I said before, we've also tried to re-install LR Agent, but results are the same.
How can we justified this difference between results?
Thanks for all the support,
What's the version of Load Generators in use?
Are there any Anti Virus or firewall enabled on the win server?
If possible, you can set up procmon to capture the activities of mmdrv.exe process on win server and send us the log. It will show all the libraries loaded into the LG process.
Also, are you running the same group of the same script on both LG?
To separate the concern, I suggest you to create a new script in VuGen, and replace only the script files like action.c, vuser_init.c, etc, and run the script only on the win server LG to see if it makes a difference or not.
thanks again for your interest on our problem and sorry for the late response; we have made some others controls and tests but the results still the same: average response time on Windows Server generators are high so we have bad performance that we can't explain.
Here attached a portion from the log extracted with PerfMon.exe application during the test on a "bad" LG machine: we have launched the web service from a Controller machine using the Windows Server as LG (1 Vusers for three minutes).
Can you take a look on it?