server performance .net agent.
Contents
20pxIntroduction
Is there any known issues for dynatrace .net agent.
Because lot of time people complain whenever the agetn is active the performance of the server and IIS goes down.+100% cpu utilization without any activity on the server.
Is this because of no of sensors?
...
Hello ALL,
I have resolved this issue.The problem was number of sensors which were selected.
The method was right for earlier versions of the dynatrace but from 5.0 the you don't have need to select an in sensors still you will some good number of purepaths.The tool itself in background have some sensors and reduce the performance overhead on the server.
You can also add the particular method from the purepath to sensors if you find it interesting.
Please let me know if any one have something to say on this.
Kevin, thanks for the feedback.
I can say for sure that the recycling happens because of the worker processes reaching the private bytes memory limit, which is currently set to around 260MB. Last time I checked this occurs at least once every 10 minutes.
We have a maximum of 6 worker processes configured for the pool and the server has a total of 16 GB physical memory. The task manager tells me there's still some free memory capacity left. Would it be a good idea to increase the private memory limit for the worker processes so we hopefully get less recycling events? Of course it would be preferable to have none at all. So, perhaps this is only a symptomatic cure for a much deeper problem (e.g. memory leak)...
On a side note I should perhaps add that my experience with IIS and .NET has been rather limited so far, therefore it's a little hard for me to recognize abnormal behavior and also what the best practices are in this context... Any input is greatly appreciated!
PS: From my (limited) understanding on how the CLR manages memory an OOM exception is rather unlikely to occur, unless the host system runs out of resources. AFAIK one cannot limit the heap size for .NET applications in the same way as it is possible for Java VMs...
I just wanted to let you guys know that I'm having similar problems, namely .NET agents with IIS 7 and minimal instrumentation (# of PP nodes <50) causing huge intermittent performance degradation on the SUD. As far as I can tell I'm not seeing overly high CPU usage, only drastic increase in response times whenever agents are injected and providing data. This is with version 4.2.0.3154. I filed a regular support ticket including all relevant information.
Anyway, not sure if the following can have an impact but I'm observing fairly frequent recycling of certain worker processes in the IIS application pool (w3wp.exe), which I suppose is normal behavior under load conditions. But isn't this potentially expensive in terms of DT agent injection and initialization?
Enrico-
There may be something else going on here too, so your inquiry into support was a good idea. Your assertion is correct though, the start up of an application will be impacted by our agent enabled, as as this is when our agents are discovering all of the code as it is loaded at runtime. The start up can take longer than usual, and the first instance of each request will also typically have a higher response time than the rest.
If the .Net application you mention is recycling as frequently as you imply, this would definitely have an impact on the application's response time.
IIS is typically configured by default to restart app pools after a certain period of time (as they are also configured to go to sleep after a period of time being idle). It has been a while since I have looked at these settings, but the default used to be something like every 22 hours. If they are restarting due to load, then there may be a resource issue involved, like the heap filling up and causing an OOM exception).
Just some thoughts.
Kevin
Hello,
Sorry for the late reply and one question.
Does this XML will carry and internal sensitive data which should not be shared on this kind of forum???
hi Swapnil,
the system profile XML will contain your configuration settings in the system profile: sensors, measures, BTs, tasks, APIs, monitors, agent groups, ... generally there should no sensitive data be in there.
you can also double-check by going through the XML, it's (mostly) human-readable.
best,
Christian
Hi Swapnil,
thanks! Could you upload the system profile you were using on that server, so I can check the settings?
Thanks,
Christoph
Hi Christoph,
How i can upload the system profile. Sorry but u need to advice me how we can do that!!!!
Thanks,
swapnil
You can export the System Profile in your dynaTrace client -> right click on the profile in the Cockpit and select "Export System Profile"
You will get an xml file that you can then upload as part of a new reply to this post
Hi Swapnil,
The .NET agent should not add that much CPU overhead. In order to help you, the following questions would be useful:
1. Which version of dynaTrace do you use?
2. Which IIS & Windows version do you use?
3. What did you change in your system profile, compared to a default profile? Certain (non-default) settings & overinstrumentation are known to cause overhead.
4. Do you have the .NET Agent Configuration Tool running on that server? In older versions it is known to use up CPU time.
5. Can you see which process is using up all CPU time?
Thanks,
Christoph
Hello Christoph,
1) Dynatrace 5.0
2) IIS 7.0 and win server 2008 R2.
3)Yes i have created system profile and mapped agent to this profile.
I also have sensor pack created for this system profile.
4) No the .net agent configuration tool is there but after the use we generally close it.
5)Need to check since de-selected the all the agent's from this particular server in .net agent configuration tool due to this issue.
Thanks,
swapnil
