Before you begin
- A Professional plan or above: the page and its API are tier-gated
- Host pools with Start VM on Connect enabled. Only those pools are eligible
- To view the page, the pool-management permission (pool.manage) or the audit-view permission (audit.view)
- To run a scan, change the optimization mode or edit schedules, the policy-push permission (pool.policy.push), held by Administrators by default
How it works
- When a session host is deallocated and has no sessions, Atlas queues its OS disk to drop from Premium SSD to Standard HDD
- When a user connects (Start VM on Connect) or an admin starts the VM, the disk moves back to its performance tier before the host is used
- Atlas keeps a small standby buffer of powered-off hosts in each pool on the performance disk, and keeps more when the pool is busy, so the next connection usually lands on a fast disk
- Only host pools with Start VM on Connect are eligible. Pools without it are left alone
Disk changes run through a rate-limited queue and are applied as ARM deployments, so a large pool is worked through over several minutes rather than all at once.
Turn it on
Open Settings → Matrix (the Settings Matrix page) and find the Smart Disk Tiering row in the Compute & VM Lifecycle category. It is off by default. Turn it on globally or for a tenant, then override it per host pool where a pool should behave differently. A pool's own setting also lets you choose the tier its disk returns to when the host starts: Premium SSD (the default) or Standard SSD.
Start with one or two pools whose users sign in at predictable times, watch the first logons for a few days, then widen it.
Optimization modes
On the Smart Disk Tiering tab, the Optimization Mode selector controls when downgrades may run:
- Automatic: downgrades happen as soon as eligible hosts deallocate
- Scheduled Only: downgrades happen only inside the schedule window you set, with a window start, a window end, the days of the week and a time zone. A window can span midnight. With no window configured, downgrades are blocked
- Manual Only: no automatic downgrades. Disks are only moved when someone runs a scan
Changing the mode asks for confirmation. Switching to Manual Only stops new downgrades until you change it back; disks that need to come back up still do.
Business hours: warm disks before people sign in
With Start VM on Connect, Azure powers the host on when a user connects, so Atlas has no chance to move the disk up first. Business hours close that gap. Tell Atlas when your people work, and during those hours every powered-off host in a Start VM on Connect pool is kept on its performance disk. When the first user of the day connects, the host boots on SSD, not HDD.
When business hours end, hosts are allowed back down to Standard HDD, so the savings continue every night and weekend. Outside business hours only the small standby buffer per pool stays on the performance disk.
Business hours are the one setting that keeps everyone off Standard HDD during the working day. If you turn on Smart Disk Tiering for pools with predictable sign-ins, set business hours too.
Set business hours in the Settings Matrix
Business hours live in the Settings Matrix (Settings → Matrix in the sidebar), in the Compute & VM Lifecycle category, next to the Smart Disk Tiering row. Search the matrix for "business hours" to jump straight to them. They need Smart Disk Tiering to be on for the same scope.
- Open Settings → Matrix and find Business Hours in the Compute & VM Lifecycle category
- Choose the scope: Global for every tenant, a tenant, or a single host pool
- Set the days, the start and end time and the time zone your users work in, then save
- Optionally set Business Hours Pre-warm Lead Time, from 0 to 240 minutes. Atlas starts warming hosts that many minutes before business hours open
- Lead time: hosts are warmed a few at a time, so a large pool needs a longer head start to be fully warm when the first user signs in. Hosts warmed early are billed at the performance-disk rate for that extra time
- Cascade: business hours flow Global → Tenant → Host Pool. A host pool in another time zone or on a night shift can keep its own hours, and a pool override with no hours opts that pool out of the warm-up
- Same value, two places: the tenant's business hours are also shown as the Daily Warm-up Window on the Smart Disk Tiering page. Both edit the same setting
- In Scheduled Only mode, downgrades after business hours still run only inside your downgrade schedule
The trade-off: first logon
When a user connects to a host whose disk is still on Standard HDD, that session runs from the slower disk until the host is next stopped, so the first logon can be slower than usual. Business hours prevent this during the working day. Leave nights and weekends to tiering, where the savings are.
What the page shows
Open Atlas Operations → Compass → Optimization and choose the Smart Disk Tiering tab (/intelligence/smart-disk-tiering). The top of the page shows Global Status with the share of pools enabled, Active Downgrades (VMs currently on Standard HDD), Monthly Savings and Lifetime Savings. Below that:
- Disk Tier Queue: pending upgrades and downgrades with each VM, direction, target tier, retries and wait time, plus how many operations the 10-minute rate limit has used. Run Scan Now is here
- ARM Deployment Jobs: each deployment Atlas ran to change disk tiers, with its resource group, disk count, direction, status and start and finish times. Hover a failed job to see the error
- Host Pool Status: each eligible pool with whether tiering is enabled, its VM, running, deallocated and session counts, and whether it uses its own setting (Pool-level) or the Global one
- Active Disk Savings: every disk currently downgraded, with its original and current tier, size, hourly and monthly savings, and since when
Run a scan
Run Scan Now checks every deallocated VM across the eligible pools in the selected tenant and queues any tier changes at the configured rate limit. Confirm the dialog and the scan usually finishes in 10–20 seconds, then shows how many pools were scanned, how many VMs were checked and how many were queued for downgrade. Only one scan can run per tenant at a time. In Manual Only mode, this is the only way disks are downgraded.