Limit on featureclasses in featureservice, performance drop.

[RobertBlok1]

We are using ArcGIS Enterprise 11.2, ArcGIS Pro 3.4, Postgress Database and a MGCP datamodel that includes more then 350 featureclasses.

All of the featureclasses are published in one featureservice. Every new feature or edit task takes about 7 seconds to finish. 
ArcGIS.Desktop.Editing.CreateSketchTool.OnFinishSketch takes most time.

If we create a featureservice with only 13 featureclasses, the performance is very good. Zero wait.

Is there an advisable number of featureclasses in a featureservice with good performance?

[George_Thompson]

That seems like a lot of layers in a service. How many instances of the service do you have running and how is the memory of the ArcSOC's on the machine hosting the service?

[RobertBlok1]

The maximum of instances is set to 20. There is no indication that we are on the max. Memory on the server is most of the time less then 60%.

[Scott_Tansley]

I have always worked on the basis that 20 is a nice number of layers in a service, 50 is still okay, 100 is questionable.  

I try and theme my services to contain the required layers and keep them short and snappy.

There is no precise answer, it's a bit of trial and error but 1 service with 350 doesn't fell right at all, sorry.

[RobertBlok1]

I published it today en kept it to a max of 50 layers per service.
I've noticed a considerable increase in performance after moving the layers outside of the featuredataset (and the max of 50 layers per service).

[Scott_Tansley]

Yeah feature datasets are not the logical container that everyone thinks they are.  They're for advanced spatial types only, and not a 'folder'.  Have a read of this:

https://esriaustraliatechblog.wordpress.com/2011/10/06/can-you-use-feature-datasets-as-logical-folde...

Its old, but still current.