Spread Sign-In Storms Across Hosts
Depth-first load balancing fills one session host to its Max users per VM before it sends anyone to the next. That keeps hosts free for autoscale to stop, but during a sign-in rush it puts every new user on the same host just when profile loading and app start-up use the most CPU and disk. Spread Sign-In Storms Across Hosts watches each depth-first host pool you turn it on for. When many users sign in within a few minutes, Atlas switches the pool to breadth-first so the rush spreads across the running hosts. Once sign-ins slow down, it switches the pool back to depth-first. The setting is off by default.
Prerequisites
- A plan that includes the autoscale engine
- A pooled host pool that uses depth-first load balancing
- An administrator who can change settings in the Configuration Matrix for the deployment, a tenant or a host pool
How Atlas detects a sign-in storm
Every minute, Atlas counts how many of the host pool's sessions started in the last five minutes. A user who reconnects to a session they already had is not counted, because Azure keeps that session's start time. When the count reaches Sign-Ins That Start a Storm (20 by default), Atlas treats it as a storm.
What Atlas changes
- During a storm, Atlas switches the pool's load balancing from depth-first to breadth-first in Azure. Each new user then goes to the running session host with the fewest sessions
- Atlas switches the pool back to depth-first once fewer than half that many sessions start in five minutes, and no sooner than 15 minutes after the switch, so a short pause in the rush does not flip the pool back and forth
- Each switch is recorded in the host pool's history and in the activity log, with how many users had signed in
- If the pool inherited its load balancing from Global or Tenant before the storm, it still inherits it afterwards
What stays the same
- Breadth-first and personal host pools are never changed
- Breadth-first only spreads users across hosts that are already running and taking new sessions. It does not start hosts: use autoscale or Start VM on Connect for that
- Users who are already signed in stay on their host
- Nothing is switched while a maintenance window that pauses autoscale is running for the pool
If you change the pool's load balancing yourself while Atlas has it on breadth-first, your choice is kept and Atlas does not switch it back.
Turn it on
- 1Open Settings → Matrix and find Spread Sign-In Storms Across Hosts.
- 2Turn it on at Global for every host pool, for one tenant, or for a single host pool.
- 3Optionally set Sign-Ins That Start a Storm for the same scope. Pick a number a normal morning does not reach: for example 10 for a 50-user pool, or 50 for a pool of 1,000 users.
Turning the setting off switches back any pool that Atlas has on breadth-first, on its next check.