Request count of Percentile 90 Response Time

[Biaggio]

Guys, I have a dashboard with some response time measurements and one of them is the 90th percentile. I have the time, OK, but I was asked how many calls are within that time, within the P90. I didn't find a way to verify this and chatting with the chat, I wasn't given a solution either. Does anyone have an idea how to extract this information?

I need to know how many calls were inside that P95:

builtin:service.keyRequest.response.server:filter(and(or(in("dt.entity.service_method",entitySelector("type(service_method),entityName.equals(~"{keyrequest}~")"))))):splitBy():percentile(95.0):auto:sort(value(percentile(95.0),descending)):limit(100)

Thank you

[michiel_otten]

Hi there,

I'm not sure if this is possible within the standard dashboarding. What you can do is create a multidimensional analysis to check it. So firstly use the query above to get the value, then use the Multidimensional Analysis to get the number of calls. 

This will give you the number.

KR.
Michiel

[Julius_Loman]

If I got your question right - I believe you have the answer already - for 90 percentile - 90% of requests are below the 90 percentile. Same for any percentile. So you can use the builtin:service.keyRequest.count.server metric for the number.  But that's for metrics which have 1-minute resolution, not for individual requests see @michiel_otten answer. - this will be the most precise answer, but you can't have that on a dashboard.

[michiel_otten · Accepted solution]

To add on Julius:  you can combine your data explorer query with metric:
(builtin:service.keyRequest.count.total:splitBy("dt.entity.service_method")*0.9) for for fast request count
(builtin:service.keyRequest.count.total:splitBy("dt.entity.service_method")*0.1) for slowest 10%

Tried playing around with percentile in metric above but does not seem to give the right number.
@Julius_Loman what's your experience with this?

[Julius_Loman]

Metric precision will never match the MDA since with metrics you work with 1-minute resolution aggregates at minimum. With MDA it's per request, and it will provide the most accurate answer.