MDA - Inaccurate results after refine

[roman_zo]

Hello,

we noticed that using refine button in MDA sometimes gives bad results.

It is a strange behaviour, as using refine button should be accurancy improved. 

For example, for simple request count metric for one service in 1 day timeframe.

• MDA results (1/64 subset) = 2M
• Refined MDA results (1/4 subset) = 1.2M
• Metric i Data Explorer = 2M

It seems it happen only on weekend and public holiday days.

Yes, MDA is giving approximate results, but the difference (50-200%) is too huge  in this case.

Is it considered as expected behaviour ?
Does anybody have similar experience in last month?

Kind regards,

Roman

[josef_schiessl]

Hello Roman!
How much "refining" the analysis improves the accuracy still depends on the amount of data that has to be read. In case not all data can be read due to performance reasons, the refine button is shown and clicking it increases the accuracy. But still there might be cases when there is high amount of data where we still won't read all of it, since it would take too long.

Possibly there are more requests in your data / analysis on weekends / public holydays?

To get 100% accurate data I would encourage you to use filters (eg.: MZ) or decrease the timeframe.

Kind regards,

Josef

[roman_zo]

Hello Josef,

thank you for replay.

no, there is less requests on weekends.

MDA is performed on single service, so I guess using MZ doesn't solve the issue.

I found that using refine button in MDA returns incorrect value for service/client request count metrics. Maybe it happens also for other metrics, but it is not so easy to be verified. It happens only in some environments, for some (high-volume) services and for some days, even in short time intervals (i.e. 2 hours)

Kind regards,

Roman

[josef_schiessl · Accepted solution]

Hello again!
As said in general its all about the amount of data that has to be read. 
If you need further guidance I would suggest to open a support case so we can have a detailed look into the data.

Thanks,

Josef

[roman_zo]

Hi Josef,

Here is answer from the support

Since the data is still based on a portion, not the whole, errors in the stimulation can and will happen. We do not influence this, unfortunately. The only option is to create a metric covering the entire range of requests. Then there is no need to use refine as the rate for data =1 (not 1/64 or 1/256 or anything else). With calc metric you will have real numbers of requests each time, each timeframe.

It seems, solving this kind of the issue for Dynatrace Managed doesn't have high priority in the Dynatrace at this moment.

with best regards,

Roman