Microsoft opened the Windows Virtual Desktop public preview on March 21, and we had a host pool serving desktops by the evening of the 22nd. We have been waiting for this since the Ignite 2018 announcement — not because we lack VDI options (we run Citrix XenDesktop and RDS estates north of 10,000 concurrent users), but because Windows 10 multi-session and the licensing entitlement change the economics of an entire class of deployment. This is our first-build write-up: what the preview actually looks like, what surprised us, and what we would put in front of a client today versus what needs to bake.
Short version for the impatient: the control plane works, multi-session Windows 10 is real and behaves better than we expected, FSLogix integration is the sleeper feature, and the management tooling is PowerShell-or-nothing in a way that will filter out casual evaluators. That last part is not entirely a criticism.
Setup notes below are from a preview environment in our lab tenant — a two-host pool in East US 2, domain-joined to an AD DS deployment we already run in Azure, with FSLogix profiles on an Azure Files share. Preview means preview: expect all of this to shift before GA.
The model: tenants, host pools, app groups
WVD's object model takes about an hour to internalize if you come from Citrix or RDS. Microsoft hosts the control plane — broker, gateway, web access, diagnostics — as a service. You bring the session hosts, the network, and the identity. The hierarchy is: a WVD tenant (bound to your Azure AD), containing host pools (collections of identically configured session host VMs), containing app groups (either a full desktop app group or RemoteApp collections) that you assign to users.
What you do not build anymore is the part of every RDS deployment that hurt: no RD Connection Broker to cluster, no RD Gateway farm, no RD Web servers, no SQL database for the broker to argue with. Anyone who has done a 2 a.m. RD broker database recovery understands what is being offered here.
Provisioning in the preview is honest about its maturity. You consent to two Azure AD applications (the WVD server and client apps), create the tenant with PowerShell, then deploy session hosts via an Azure Marketplace template:
Import-Module Microsoft.RDInfra.RDPowerShell
Add-RdsAccount -DeploymentUrl "https://rdbroker.wvd.microsoft.com"
New-RdsTenant -Name "HMT-Lab" -AadTenantId $aadId -AzureSubscriptionId $subId
New-RdsAppGroup -TenantName "HMT-Lab" -HostPoolName "pool01" -Name "Desktop"
There is no portal experience for the WVD objects themselves — host pools and app groups do not appear as Azure resources. Everything is the Microsoft.RDInfra.RDPowerShell module plus a management web UI you deploy yourself from a template if you want one. For a preview, fine. For the enterprise operations teams we work with, the PowerShell-first posture is actually workable — our RDS and Citrix runbooks are already scripted — but the lack of ARM-native objects means no tags, no RBAC granularity on pools, and no policy. Noted for the GA wishlist.
Windows 10 multi-session: the headline feature holds up
The technical novelty here is Windows 10 Enterprise for Virtual Desktops — a Windows 10 SKU that accepts multiple interactive sessions like a Server OS, while presenting a genuine Windows 10 experience to the user. This matters because the RDSH compromise we have lived with for a decade — Server 2016 wearing a desktop-experience costume — breaks a meaningful set of applications and confuses users with "Windows Server" branding in odd places.
We loaded a pool with our standard knowledge-worker app stack: Office 365 ProPlus (which, notably, is properly supported on multi-session — another thing Server-based RDSH never got), a line-of-business Win32 app, Chrome, and Teams. Everything installed and behaved as it would on physical Windows 10 1809. App compatibility checks that sniff for a workstation SKU pass. Per-user MSI behavior worked where the same package needed shims on RDSH.
Density is the question everyone asks. Early impression from synthetic loads: comparable to Server 2016 RDSH on the same VM size — we ran a D8s v3 with a dozen simulated task-worker sessions comfortably. We are not publishing users-per-vCPU ratios from a preview lab; we will do that from production telemetry when it exists. Treat vendor density claims — anyone's — as marketing until measured on your workload.
The licensing story remains the Ignite bombshell it was: if your users have Microsoft 365 E3/E5 or Windows E3/E5 entitlements, WVD access rights are included. You pay for the Azure compute, storage, and networking, not for a per-user VDI platform fee. Run that math against a Citrix renewal at enterprise scale and you understand why we cleared lab time the week the preview opened.
FSLogix is the part that makes it production-shaped
Microsoft bought FSLogix in November and the reasoning is visible here. Multi-session, pooled desktops live or die on profile handling, and roaming profiles plus folder redirection was never a real answer — anyone who has watched Outlook rebuild an OST on every logon knows the failure mode. FSLogix Profile Container mounts the whole profile as a VHD from an SMB share at logon. The user gets a local-feeling profile; Outlook cached mode, OneDrive sync state, and search indexes persist across session hosts.
In the preview lab we pointed containers at Azure Files. Configuration is a handful of registry values on the golden image (Enabled, VHDLocations) and an AD group. Logon-to-desktop with a warm container was consistently reasonable in our tests, and — the part that matters — the second logon on a different host behaved identically. That is the property that makes pooled non-persistent desktops viable for Office 365 workloads.
Caveats we hit: watch the Azure Files performance tier for anything beyond a lab (standard file shares will bottleneck a real pool; we want to test premium files before recommending), and plan container sizing deliberately — default max VHD sizes are generous and a few hundred users of default-sized containers is real storage money.
The gaps versus Citrix, cataloged honestly
We run XenDesktop 7.15 LTSR estates for a living, so here is the sober comparison. What WVD preview does not have:
- Image management. There is no MCS or PVS equivalent. You build a golden image, then you update a host pool by deploying new VMs from the new image and draining the old ones. Scriptable, but it is your script.
- Autoscale. No native power management of session hosts against demand. At on-peak/off-peak enterprise scale this is where the Azure bill is won or lost. Also scriptable via Automation; also your script.
- Advanced graphics and peripheral remoting. HDX is a decade of protocol engineering. RDP has improved enormously, but for 3D CAD, medical imaging, or aggressive USB redirection, Citrix keeps its lead today.
- Monitoring. Diagnostics exist as a service API in preview; there is no Director equivalent. Helpdesk workflows need building.
- App layering, session recording, granular policy — the long tail of platform features enterprises lean on.
Citrix has also announced it will manage WVD workloads and multi-session Windows 10 from its own control plane, which is the pragmatic acknowledgment that the OS entitlement changed the game: the interesting future fight is control-plane value, not OS access rights.
What we would deploy today, and what waits
Deploy today (well — pilot today, deploy at GA): task-worker and knowledge-worker pooled desktops where the app stack is Office-plus-browser-plus-a-few-Win32-apps; RemoteApp publishing of legacy line-of-business apps; and — the scenario we think is underrated — burst and contractor capacity attached to an existing estate, where "no platform licensing, pay for compute by the hour" is precisely the right shape.
Wait on: anything graphics-intensive, anything with exotic peripherals, estates that need mature day-2 operations tooling out of the box, and — obviously — anything a client would call production, because this is a public preview with no SLA. We are also watching Windows 7 support in WVD with interest, since free extended security updates for Win7 VMs in WVD were part of the announcement, and every enterprise we serve has a Windows 7 long tail with January 2020 approaching.
Our plan between now and GA: build the autoscale and image-rotation scripting we will clearly need, benchmark FSLogix on premium file storage, and load-test density against our real Citrix baselines rather than lab synthetics. When Microsoft puts an SLA on the service, we intend to have opinions backed by numbers.
If you're facing this
If you run RDS or Citrix today and the WVD announcement is on your 2019 evaluation list, we have a working preview environment and the beginnings of a real cost model. Get in touch and we will share what a pilot for your workload profile would look like — including the parts where the honest answer is "not yet."