We have about 600 map services in a multi-machine AGS site. To keep the number of SOCs low we have the minimum instance set to 0 and max as 2, for most of these services. When the webmaps containing these layers with 0 min instances are accessed in the map viewer, the layers fail to load, and eventually timeout.
The AGS server logs, simply state failed to create an instance, even though the validation of the registered db passes. If we set the minimum instance to 1, then the services load without any issues. I am starting to think the server is failing to spin up new ArcSOCs when it has a large number of services published.
Interestingly, these services performed better in 10.8.1 server setup, we had to keep validating the database every few hours though to make it work in 10.8.1, and doing the same doesn't seem to help in 11.1
Any inputs from community members and Esri staff would be helpful.