Does Dynatrace agent cause any overhead on the response time values?

[GauriS]

I recently executed a performance test and was analyzing the results using Dynatrace. I was curious to know if Dynatrace agents would have caused any overhead on the response time values. 

Is there any official explanation or documentation on this topic?

I would appreciate if I can get a proper explanation on whether Dynatrace agents would cause any overhead on the response time values.

Thank you.

[ChadTurner]

Hello, Dynatrace doesn't add any overhead to the underlying infrastructure and there are safeguards in place if the resource consumption rises above the threshold that dynatrace has coded in. Is someone at your Org reporting overhead with the Agent? 

[GauriS]

Thanks Chad for your quick response.

The management is asking to execute the whole performance test again by uninstalling the agents as they think the agents might have caused an overhead and eventually impacted the response time values.

Is there any documentation supporting this which can be presented to the folks?

Thanks! 

[ChadTurner · Accepted solution]

Before you turn off Dynatrace I recommend taking a look at the App and the underlying services within Dynatrace as that should point out the issue with the app. If you remove Dynatrace you remove the ability to determine what is causing overhead. The closest I would get to removing the host is downgrading it to 1/2 stack so you still have monitoring but no RUM Injection. 

Sadly I've had this come up more then id like it to and each and every time it boils down to poorly coded apps. I burned an entire year onboarding a set of hosts for an app based off the same claim of the OneAgent introducing performance issues. 

[GauriS]

Thank you Chad.

[AntonioSousa]

@GauriS,

I have also been asked this by many clients and prospects.

The overhead of Dynatrace is very low, but it does exist. One of the good things about Dynatrace is that OneAgent monitors what's going on in the servers where it is installed, and so it measures what it consumes itself. This includes CPU, memory and bandwidth, besides the space it occupies on disk.

What I have seen is that resource consumption is very low, and only in one specific case, did it go above expectations. What I normally see is that overhead is even "negative", in the sense that when you use it, if you improve given on the feedback that you receive, the resource consumption will be lower than before you installed it. An example:

• You install OneAgent on a server that has regular 70% CPU consumption.
• You verify that you are able to improve CPU consumption because of badly coded apps, as @ChadTurner mentioned. You improve, say 10% of CPU consumption. You can eventually even reduce more.
• You verify that OneAgent doesn'y go above 1% CPU consumption.
• With Dynatrace, the overhead will be going from 70% regular usage to 61% usage. Hence, a -9% overhead 😂

I have had many cases where the above applies 😎

[Romanenkov_Al3x]

This is a good approach, but not for large Customers where more than 30 people on every position include perfomance engineers.

CPU/Memory/Storage is not overhead at all if Customer can add x2 resources without problems. 

The interesting cases when you have added +16vcpu, +64GB mem, and still have Response time overhead 😉

Regards,

Romnanenkov Alex

[AntonioSousa]

@Romanenkov_Al3x, I have also had cases where more resources, with same load, lead to degradation. Not usual, but I have seen it happen...

[Romanenkov_Al3x]

That's a great point. I will try this in future tickets.