User Profile Disks are the built-in answer to the roaming profile problem on a Remote Desktop Session Host: instead of copying a profile across the network at every logon, Windows mounts a VHDX that holds the user’s profile and leaves it on a share. On a single Windows VPS running RDS it is the cheapest way to give ten or twenty users a consistent desktop, and it is also the source of the most common “the server filled up overnight” support call.
This guide covers the UPD share layout, the size and placement decisions that matter, how to check what is actually consuming space inside each VHDX, and how to reclaim space without corrupting a live session.
How User Profile Disks actually behave
When UPD is enabled for a collection, the RDS host intercepts the profile at logon and mounts UVHD-template.vhdx for new users, or a per-user copy named with a SID hash. Only the profile lives in the VHDX. The user’s home drive, redirected folders and any data outside the profile path stay where they are, which is why UPD and folder redirection are usually deployed together: put the bulky data on a plain file share and keep the disk small.
The VHDX comes in two modes. Dynamic disks grow as needed and are the default for a reason — most users use a fraction of their allocated size. Fixed disks reserve the full size up front, which makes capacity planning trivial and eliminates the “disk full inside the VHD” failure that dynamic disks produce when a session writes past its limit. On a small VPS where the storage back end is thin-provisioned anyway, fixed disks do not actually cost you more physical space than the dynamic growth you would have to accommodate regardless.
Share layout and permissions
Create one share for the disks and a separate path for the template. The share ACL grants Full Control to the RDS computer account (or the security group holding all session hosts), not to individual users. Users must never be able to browse each other’s VHDX files — mounting someone else’s profile disk is a trivial privilege escalation.
A working layout:
D:\UPD\
Collection1\ <- per-collection folder
UVHD-template.vhdx
UVHD-<SID-hash>.vhdx
New-Item -ItemType Directory -Path D:\UPD\Collection1
New-SmbShare -Name UPD$ -Path D:\UPD -FullAccess "CORP\RDS-Hosts$"
Put the UPD share on a separate volume from the system disk, so a runaway profile disk cannot take the OS partition down with it. If the VPS has a single data disk, this is one more argument for adding a second data disk before you enable UPD rather than after.
Sizing the disks and the volume
The default template size of 20 GB is fine for Office and a browser, and too small for anyone installing Visual Studio or a local database. Do not size the disk for the worst case per user; size it for the typical case and monitor. A useful starting point for a small VPS deployment:
- Disk size: 20 GB dynamic for general office users, 40 GB if the profile will hold development tools.
- Volume headroom: keep at least 30 percent of the UPD volume free. Dynamic VHDX files fragment badly when the volume is above 85 percent utilisation.
- Per-disk allocation: set the collection’s maximum disk size in Server Manager under Collection Properties, User Profile Disks.
Ten users at 20 GB plus 30 percent headroom means about 260 GB of volume. On a VPS where the data disk is 100 GB, either reduce the per-disk maximum to 10 GB and redirect Documents and Desktop to a file share, or move to FSLogix containers, which handle Office cache data in a single file more efficiently than a whole profile VHDX.
Diagnosing growth: what is inside the VHDX
The question is never “why is this disk full” but “which files filled it”. Mount a suspect disk read-only on a maintenance window, or copy it to a scratch path first if the user is logged on. The session host itself will not let you double-mount it.
# Copy the suspect disk, then mount it read-only
Copy-Item "D:\UPD\Collection1\UVHD-*.vhdx" C:\temp\probe.vhdx
Mount-VHD -Path C:\temp\probe.vhdx -ReadOnly
Get-Volume | Where-Object DriveLetter -eq 'F'
# Inspect the mounted profile
Get-ChildItem F:\Users\* -Recurse -Force -ErrorAction SilentlyContinue |
Sort-Object Length -Descending | Select-Object -First 30 FullName,Length
The usual culprits, in order: AppData\Local\Microsoft\Outlook with an OST file for a mailbox far larger than expected; browser caches under AppData\Local\Google or AppData\Local\Microsoft\Edge; AppData\Local\Temp full of installer debris; and the Windows Search index database. None of these belong in a roaming profile at all, and each has a policy fix.
Exchange cache mode is the biggest single lever. Setting the cached exchange mode slider to 1 GB or 3 months for RDS users keeps the OST bounded, and is standard practice for any hosted desktop. For browser caches, an Active Directory Group Policy that redirects the cache directory to a local C:\Users\Public\Cache path removes it from the profile entirely.
Reclaiming space without breaking a session
A dynamic VHDX does not shrink when files are deleted inside it — the blocks stay allocated. Compaction requires the disk to be offline, which means the user cannot be logged on. The safe sequence is: sign the user out, verify no session is holding the disk, confirm the disk is not mounted, then compact.
# Confirm the disk is not in use by any process
Get-Process | Where-Object { $_.Modules.FileName -like '*UVHD-*' }
# Compact after the user is signed out
Optimize-VHD -Path "D:\UPD\Collection1\UVHD-<hash>.vhdx" -Mode Full
If Optimize-VHD fails with “the file is in use”, a disconnected session is probably still holding it. RDS keeps disconnected sessions alive for the configured idle timeout, and a disconnected session still owns its profile disk. Either force logoff the session in Server Manager or use logoff against the session ID from query session. This is also why the idle timeout setting for the collection matters for disk hygiene, not just licence conservation.
Backup, corruption and the graceful fallback
UPD disks are ordinary VHDX files and can be backed up by any VSS-aware file-level backup, as long as the user is signed out or the backup product supports open-file snapshots. A volume shadow copy of the UPD share taken mid-session can capture a disk that is being written, and restoring it lands the user in a half-committed state. Schedule backups of the UPD volume outside working hours and verify restore on a test user before you need it.
When a disk corrupts, the recovery is unglamorous but reliable: rename the corrupted VHDX, let the user log on to create a fresh one, then mount the old disk read-only and copy back the files they care about from the Desktop, Documents and Downloads directories. Do not attempt to repair a mounted profile disk with chkdsk — repairing a VHDX that Windows has mounted as a live volume risks turning one corrupt profile into a corrupt share.
When UPD is the wrong choice
UPD does not support multiple session hosts for the same collection in a shared configuration — each host needs its own collection and its own share, or you move to FSLogix. It also struggles with Office 365’s modern authentication caches and with any application that stores large local caches outside the profile path. If the workload involves Teams, OneDrive or a heavy IDE, budget the time to evaluate FSLogix profile containers rather than fighting the VHDX growth curve.
For a small team on a single Windows VPS with a handful of line-of-business apps, though, UPD with a disciplined share layout, a 1 GB Outlook cache and redirected user folders is stable, free and easy to explain to whoever inherits the server. Measure the volume monthly, compact during maintenance windows, and it will not surprise you.
UPD growth is really a disk-provisioning question, and the answer is usually a second attached volume rather than a bigger system disk. The Windows VPS plans list how many data disks each tier accepts and their sizes, which is the number that decides how many users you can realistically host on one machine.


