What is the correct configuration for a differential control to avoid that an average response time metric affects the calculation when its value falls very close to zero, which causes a high behavior change generating an alert for differential control.
Here is some information on the Differential Controls:
Configure Differential Analysis - CA Application Performance Management - 10.5 - CA Technologies Documentation
Do you know if an alert state change or variation in the metric value is triggering notifications?
Just to clarify the question...in what scenario would it be a special case for differential control when average response time is close to zero and when that changes?
I see no indication if the information was helpful or not. No further questions were asked. Marking as answered. However, additional questions may be asked as needed.
Thanks for the response by Thomas.
My question is about this
1.- The average response time in this case has values between 100-150 ms. This is normal for this metric.2. The DA metric for the average response time metric gives me an equal stability value of 10, however, if the average response time approaches 0. DA detects the behavior change and returns a value of> 10 and depending on how close to zero it can be> 20 or> 30.
Objectively: an average response time close to 0, should consider it good and DA should give me a value <= 10.How do I configure DC so that DA gives me a value that does not generate false positives in my alert.By having an alert based on the result of DA this generates a danger alert when in reality the monitored system is operating normally
In that case, I would try raising the caution and danger thresholds. This should keep the variance intensity metric lower overall (but you'll need to some testing). You might want to create a differential control just for these metrics as well.
DA values work like this
0 – No data10 – Everything normalAnything greater than that till 40 indicates varying degrees of breach.
So, you will never have value less than 10 unless its no-data (count is 0) in original metric
From the images attached, I am not able to see where ART is dropping to zero.
The first image indicates ART pretty stable with one spike which is reflected by one tiny bump there in variance intensity. DA is quite good in catching up to latest values so if your application is varying between 100 and close to zero and it is expected behavior eventually it will catch up, but depending on how often these two ranges are used it may take a while (since deviation and variance is affected by values in distribution)
This is what I can guess is happening. ART normally is always 100, no major deviation so it’s very less. Sometimes ART drops to 10 or stay there and DA learns in few cycles and now the prediction is around the same value. Since the values are not deviating much (yet), our band for triggering will be very close to around 10, like 15 or so. So, when it climbs back to normal 100 again, this is a change for DA now since its higher than what was expected so triggers alert. Over time if we have a healthy combination of 10, and 100 values the deviation is going to grow little higher to cover both ranges hopefully.
This may not help much so you can configure the control to act little more conservative in triggers. To do this you can modify some knobs in DC.You can increase caution and danger threshold from current values to higher values by moving sliders right. This means we will wait for more breaches before increasing variance intensity. The more right we go, the more conservative we get.
Other controls in advanced could help; but let’s come back to it only when needed.This page in docops should give reasonable idea on what each control doeshttps://docops.ca.com/ca-apm/10-5/en/administrating/manage-metric-data-by-using-management-modules/configure-differential-analysis
Hope it helps
Thanks Sergio for your response.
Originally, those who implemented DA, only defined a single DC, for this reason today it is complex to perform tests, since it affects all DAs on ART, RT and interval Defect metrics. Within the metrics of ART we have services with very even response times and others that have a great fluctuation.
Example on previous point to 13:00 (Time range: 6 hours - Resolution: 6minutes)
(Custom range: 10 min - Resolution: 15seconds)
DA can be applied on any metric?
Cuándo es correcto aplicar DA a una metrica?
Was Sergio's note helpful .Or do you have any followup questions after reviewing this response.