Context
The organization ran four VMware vSphere clusters with vSAN storage, stretched across two data centers: Windows servers, Linux servers and two clusters for Omnissa Horizon virtual desktops. That meant more than 450 server VMs plus the desktops of over a thousand users. Rising VMware licensing costs made a platform change necessary, with two constraints: no new servers to buy and no service interruptions.
I wrote the design document, from the workshops with the customer’s technical teams to the final version that guides deployment and migration. The detailed Horizon design belongs to Omnissa Professional Services; I designed the platform, the file services and the integrations between Horizon and Nutanix.
The challenge
- Off-list hardware. The nodes were vSAN Ready Nodes missing from the Nutanix hardware list, with unsupported NICs and cache drives. They needed a dedicated qualification and physical work on every node.
- No spare nodes. Each new cluster could only be built from nodes freed up by a VMware cluster, so the order of the work decided capacity, risk and timing.
- Zero RPO on two active sites. The customer asked for synchronous replication for every workload, with both data centers in production.
- No Internet access. Upgrades, images and licenses had to get into a closed perimeter.
- A mixed VM estate. Roughly one VM in nine ran an unsupported operating system, from Windows Server 2003 to RHEL 3, and more than a hundred were on extended support or could not be identified.
The approach: Eject, Build, Expand
I turned the constraint into a method. Every node goes through the same cycle: evacuate the workloads, eject it from the VMware cluster, rework the hardware (Intel E810 25 GbE NICs, cache drives removed, dedicated BIOS settings), reimage it with Foundation, then create or expand an AHV cluster. The customer’s staff does the physical work, following Nutanix procedures and with Nutanix sign-off.
The sequence starts with two bridge clusters, one per site, carved out of the VDI nodes after consolidating the desktops on a single cluster. The Windows servers move there temporarily; the nodes they free up build the Linux clusters, then the Windows clusters, then the VDI clusters. In the end every node reaches its final place, including two small test clusters for application and firmware upgrades.
The wave plan has nine steps. Each one comes with a capacity plan to validate before moving on, because N+1 headroom is the weakest point during the intermediate phases.
Key decisions
- One cluster per site, never stretched. Each group has two twin clusters that protect each other with Nutanix Disaster Recovery: asynchronous or NearSync replication as the baseline, synchronous with a witness where zero RPO is needed.
- Four Prism Central instances. Scale-out, split by servers and desktops and by site, paired as availability zones. The Horizon capacity provider depends on Prism Central, so its resilience is a requirement, not an option.
- Trade-offs on record. RF2 instead of RF3 because of capacity limits, and erasure coding on the server containers, are two customer requests. The design records them with their impact and the customer’s risk acceptance instead of hiding them.
- Active-active desktops. Pools sized at 100% on each site, with synchronous replication only for the master images. Profiles stay on the existing NAS in the first phase and move to Nutanix Files with Smart DR after functional validation.
- Microsegmentation instead of NSX. Flow Network Security isolates desktops from each other: each one talks only to the gateway. Policies start in monitor mode.
- Dark site. An internal LCM repository, offline images and licenses, Pulse turned off, and log collection from a bridge workstation controlled by the customer.
- A plan for legacy VMs. A test on an isolated network and a rollback for every VM. As a fallback, a VMware lab cluster with a deadline for each VM; nested Hyper-V only as a last resort, with the lack of support accepted in writing.
Project status
The design was delivered in October 2026 and is the reference for deployment, migration with Nutanix Move and validation testing. Decommissioning the VMware environment and the backup policies remain with the customer.
01
Solution, layer by layer
Pick a layer involved to read what was done.
L1 · Secure access
The Unified Access Gateways become native AHV appliances in the DMZ, with anti-affinity rules. A virtual ADC load-balances Connection Servers, Unified Access Gateways and App Volumes Managers, with persistence aligned to the Horizon session timeout.
L2 · Digital workspace
Omnissa Horizon on AHV through a capacity provider in Prism Central: templates, categories and compute profiles with vTPM for Windows 11. The management block runs on the server clusters, separate from the desktops. App Volumes sits on Nutanix Files from the first phase; FSLogix profiles and DEM data follow after validation.
L3 · Hyperconverged platform
Eight AHV clusters in four groups (Windows servers, Linux servers, desktops and test), one cluster per site in each group. RF2 with N+1 headroom, inline compression, four scale-out Prism Central instances, Flow Network Security on every cluster and Nutanix Files with Smart DR between the sites.
M · Platform migrations
An Eject, Build, Expand sequence in nine steps, with two temporary bridge clusters. Nutanix Move for server VMs, fresh deployments for appliances and gateways, new pools for desktops. For every wave: cut-over rules, freeze windows, rollback to the source VM and acceptance criteria.
02
Results
- 8
- Nutanix AHV clusters, four per site
- 80+
- nodes reused from the vSAN clusters
- 450+
- server VMs to migrate
- 1,000+
- virtual desktop users
03