Improve App Service Plan auto-scaling rules
As we know, auto-scale can quite easily enter an oscillation known as "flapping". This is due to the linear nature of basic autoscale rules. The use of a cooldown period helps reduce the impact of this behaviour, but I would think that we could implement better control algorithms.
I'm not quite sure what the best approach would be, but a few thoughts come to mind:
In mechanical control applications, we see PID loops. Perhaps something similar could be implemented on the aggregate load data?
Scaling thresholds based on the current instance count might work?