We have dynatrace appmon in production with ENORMOUS traffic (90k purepaths/minute) with xLarge Setup.
I have noticed that in high load situations the dynatrace starts skipping purepath (Buffer full or Dynatrace server bottleneck).
I have noticed that when we unchecked the "Failure rate calculation" in each BT (600), the purepath skipping improved a lot !
So, I am on aggressive optimization and I was looking if I can tweak anything in Measure > "Details" Tab.
Can anyone please suggest if there more parameters I can look for tuning?
At a high level it can just come down to "do I really need a measure for this?" When you're creating business transactions sometimes you aren't interested in long term analysis so you can just uncheck the box to store measures in the PW (or either of the default result measures).
For normal measures you have the option to create a measure for each application, agent, agent group etc... and when you are configuring these really take a second to think about if you would have need to chart that measure by agent, or if you're really interested in the measure for all applications or just for specific ones.
Touching on BTs again if a simple change like that is having a big impact on server performance consider reviewing those to see if you can change any configurations or remove any BTs that aren't being used any more. It could be you have some BTs with lots of splittings and the failure rate would need to be calculated for all of them which could magnify the impact as opposed to just calculating the failure rate across the board.
The Measure Health dashlet can also be helpful to see if there are any rogue BTs or system profiles that are creating a lot of measures.
Hi James, thanks for the hints. Also, If I create copy of system profile, say in case of DR profile for same application, will the newly created profile add measurement overhead too (even when DR is down)?
I'm not 100% sure if there will be any impact on the server in the case that there is no data going through that would create measures (i.e the profile or application is inactive) but I believe the real performance impact on the server is when it actually has to parse through the data it is receiving and actually process and create the measure data. So if there is no data and no measures actually being created I wouldn't think it would be significant.