Update operator-best-practices-run-at-scale.md

Adding a couple of callouts to the page
1. ephemeral OS disks for system pools
2. Stop and Start feature not available at >1k scale

Author: pavneeta

articles/aks/operator-best-practices-run-at-scale.md

@@ -39,11 +39,14 @@ To increase the node limit beyond 1000, you must have the following pre-requisit
 
 ## Node pool scaling considerations and best practices
 
-* For system node pools, use the *Standard_D16ds_v5* SKU or equivalent core/memory VM SKUs to provide sufficient compute resources for *kube-system* pods.
+* For system node pools, use the *Standard_D16ds_v5* SKU or equivalent core/memory VM SKUs with ephemeral OS disks to provide sufficient compute resources for *kube-system* pods.
 * Create at-least five user node pools to scale up to 5,000 nodes since there's a 1000 nodes per node pool limit.
 * Use cluster autoscaler wherever possible when running at-scale AKS clusters to ensure dynamic scaling of node pools based on the demand for compute resources.
 * When scaling beyond 1000 nodes without cluster autoscaler, it's recommended to scale in batches of a maximum 500 to 700 nodes at a time. These scaling operations should also have 2 mins to 5-mins sleep time between consecutive scale-ups to prevent Azure API throttling.
 
+> [!NOTE] 
+> You can't use [Stop and Start feature][Stop and Start feature] on clusters enabled with the greater than 1000 node limit
+
 ## Cluster upgrade best practices
 
 * AKS clusters have a hard limit of 5000 nodes. This limit prevents clusters from upgrading that are running at this limit since there's no more capacity do a rolling update with the max surge property. We recommend scaling the cluster down below 3000 nodes before doing cluster upgrades to provide extra capacity for node churn and minimize control plane load.
@@ -61,3 +64,4 @@ To increase the node limit beyond 1000, you must have the following pre-requisit
 <!-- LINKS - Internal -->
 [quotas-skus-regions]: quotas-skus-regions.md
 [cluster upgrades]: upgrade-cluster.md
+[Stop and Start feature]: start-stop-cluster.md