Leaving VMware puts a staffing decision in front of the platform decision. Which VMware skills carry over to OpenStack, what the team has to learn and in what order, who owns what after the move, and when hiring or managed operations is the better call.
Leaving VMware is usually budgeted as a licence and migration project. Whether it works depends on people. Someone has to run the new platform from the first production wave onward, and the team running vSphere today learned a different stack.
Two dates set the pace:
The work before the first wave also needs people. For an organization running 2,000 or more VMs on at least 100 servers, Gartner research reported by The Register puts a migration at 18 to 48 months. Its estimate also includes seven to ten full-time staff for a month of scoping and six more for up to nine months of technical evaluation. Unless the plan pays for outside help, those people come out of the team that keeps production running.
This post covers staffing a move to OpenStack. OpenStack is the open source cloud platform that runs virtual machines on Linux servers with the KVM hypervisor, usually on top of Ceph, a distributed storage system. Most plans staff it with a mix of three routes:
- Retrain the current VMware team.
- Hire engineers who have run OpenStack before.
- Buy operations: a provider builds the cloud and either runs part of it as a managed service or backs your team with support.
The decision is the mix, layer by layer. Get it wrong and the cut-over is run by engineers still learning the platform, senior VMware staff leave mid-project, or a vacancy for a scarce skill holds up the first wave.
Our take: retrain first, buy the hardest layers
Retrain the team you have for everything that touches workloads, design and process. Ceph and the control plane have the steepest learning curve. Buy the cloud build, and keep an operations backstop for those layers until your engineers pass a readiness test on each one. The test is set out in the rules at the end. Hire only for a gap that retraining cannot close before the first wave. That gap is usually one senior Linux or Ceph engineer.
Four reasons sit behind that mix:
- Time to competence is uneven. A VMware engineer can map a host cluster onto OpenStack compute early on, because the design questions are the same. Recovering a degraded Ceph cluster takes practice on a running one. Training covers the first kind of skill, and a backstop covers the second while the practice happens.
- Retention. The engineers who know which VMs matter, why a VLAN exists and whom to call when a database host fails hold knowledge no new hire brings. A plan that brings in outsiders to run the new platform tells them where their careers end, and they can take that knowledge to an employer still running VMware.
- The hiring pool is hard to size. The OpenInfra Foundation's Certified OpenStack Administrator (COA) exam, the only professional certification it supported, was passed by 128 people in 2023. That figure counts one year's certifications. It does not measure the workforce, so it cannot tell you how long a requisition will stay open.
- Cut-over risk. The first production waves are the period when your team knows the new platform least. A backstop that answers during a cut-over covers that period, and a retrained team covers everything after it.
VMware skills that carry over to OpenStack
Much of a VMware team's work needs little new learning on OpenStack. Each area below names the one thing that changes, so the training plan pays only for that.
- Capacity planning: the least new learning. Host sizing, overcommit and failure headroom are the same decisions. Nova, OpenStack's compute service, sets overcommit with CPU and RAM allocation ratios and groups hosts into host aggregates.
- Storage design: same decisions, new failure model. Tiering, failure domains and the trade-off between replication and usable capacity carry over. Ceph spreads data across disks in many servers, as vSAN does across a cluster's hosts, so vSAN experience carries further than SAN experience. Recovering Ceph is the skill that needs practice (step 5 below).
- Network design: the design work transfers. VLANs, segmentation and firewall policy carry over. NSX policies have to be mapped to security groups in Neutron, OpenStack's networking service, and validated.
- Change control: same process, busier calendar. Your change board, maintenance windows and rollback plans stay. OpenStack ships two releases a year, so budget upgrade windows on that cycle.
- Incident handling: a new escalation path. On-call rotas, severity levels and post-incident reviews carry over unchanged. Escalation moves from Broadcom support to your provider and the upstream projects.
- Vendor patch cadence: new sources to watch. The habit of tracking advisories and booking patch windows carries over. The feeds change to OpenStack Security Advisories, Ceph releases and your Linux distribution's updates.
Product muscle memory stays behind: vCenter screens, ESXi host troubleshooting and PowerCLI scripts. Replacing it is the learning plan.
OpenStack training for VMware admins, in order
What to learn first
The order follows where incidents will land after cut-over and how long each skill takes to build. It assumes a KVM deployment with VM disks on Ceph, the most common storage backend for OpenStack.
- Linux and KVM troubleshooting. In this design, every compute host is a Linux server running KVM and QEMU through libvirt. This skill is needed from the first pilot. When a VM will not boot or a disk hangs, the cause can also be in an OpenStack service or the storage backend.
- OpenStack's model and vocabulary. Projects, quotas, flavors, images and the API. The OpenInfra community's VMware to OpenStack Migration Guide, published in June 2025, compares VMware features with their OpenStack equivalents and notes that most perceived differences come from terminology.
- Neutron networking with Open vSwitch or OVN. The network team learns how virtual routers, security groups and provider networks sit on the physical network they already run.
- Ansible automation. PowerCLI scripts and orchestration workflows get rewritten. Ansible has modules for OpenStack, so the team's scripting habit carries into a new tool.
- Ceph. Storage engineers learn placement groups, disk failure and recovery. A mistake here touches the VM disks on that cluster, so the backstop stays longest on this layer.
- Kubernetes for the control plane. In Atmosphere, our OpenStack platform, the control plane runs on Kubernetes. When a provider owns this layer, we recommend learning it last and building basic Kubernetes competence before your team owns it alone.
Material and milestones
- Certification as a checkpoint. In 2023, ten OpenInfra member companies offered training for the COA.
- The code that runs your cloud. Atmosphere is open source and public on GitHub, with its deployment, upgrade and troubleshooting guides, so your engineers can study the platform they will run before it exists.
- A production-shaped test. A proof of concept that fails a node and an upgrade on purpose doubles as the team's first practice on recovery.
The first installation is a one-off project the team will rarely repeat, while operating the cloud is daily work for years. Buying the build lets the team spend its training budget on operations.
Who owns what after the VMware to OpenStack move
Write the ownership down before the first wave, layer by layer:
- VM catalogue and workload owners (images, flavors, project quotas): your current VMware administrators, from day one.
- Physical network and provider networks: your network team, from day one.
- Ceph storage: your storage engineers, with the provider as backstop until they pass the readiness test below.
- Control plane and upgrades: the provider at first, moving to your team once it passes the same test.
- Identity and access: who grants and revokes cloud accounts and roles, and who approves them.
- Backup and restore: who owns the backups of VM data and of the platform itself, and who has rehearsed a restore.
- Change control and incident command: yours in every model. The provider works inside your process.
- Monitoring and paging: decide on day one, layer by layer, whose phone each alert rings: your on-call engineer's or the provider's.
We run engagements at both ends of that split. In a support-only engagement, your team executes and our engineers act as the backstop. With full remote operations, we run the platform for you. Our professional services team designs and builds the cloud, and later checks an architecture or an upgrade plan with your engineers.
Hire or buy managed OpenStack instead?
Some teams should hire or buy operations for most of the stack. The strongest cases:
- The team is already at capacity. When retraining would fall on the same two or three people keeping VMware production patched, training time has to come from somewhere. Buy platform operations and let them learn on a cloud someone else keeps healthy.
- The date falls inside the learning curve. vSphere 8 general support ends on October 11, 2027. An estate that starts scoping late in 2026 faces the month of scoping and up to nine months of evaluation cited above for a 2,000-VM estate. That leaves little room for a team to learn before the first wave.
- The engineers want to stay on VMware. Some specialists will prefer a VMware role elsewhere, and no training plan changes that. Hiring is the honest answer for the seats they leave.
- Infrastructure is not where the business competes. If running a cloud will never be core work, fully managed operations with your team owning the workloads can be the permanent model. For small estates, OpenStack may be the wrong fit altogether.
Our take also assumes an estate big enough to justify an in-house platform team and engineers willing to learn Linux at depth. Where either is missing, the mix moves toward managed operations, and that is a sound plan.
A VMware to OpenStack staffing plan, layer by layer
Take each layer of the new cloud through these rules, in order:
- Workloads, network design, change control and incident command stay with the current team and its retraining plan in every case.
- The build, the control plane and Ceph go to a provider until your team passes a readiness test on each: it has handled troubleshooting, a recovery and an upgrade rehearsal in a representative environment, with on-call coverage in place.
- One senior hire is justified when a single skill, usually Linux and KVM depth or Ceph, has no internal candidate and the backstop would otherwise become permanent.
- Full managed operations fit when the people who would learn OpenStack are the same people still running VMware production, or when they have said they will not move.
- Move a layer in-house only when your team passes that test on it. Until then, keep the backstop and review the mix after each upgrade.
Our CEO has chaired the OpenStack Technical Committee and sits on the OpenInfra Foundation board. Our engineers operate OpenStack clouds and contribute to the upstream projects your team is learning.
Atmosphere's code and its deployment, upgrade and troubleshooting guides are public, so what your engineers learn on our build stays with them. Our post-migration support includes training and knowledge transfer for your staff.
Your team's current skills and your first-wave date are enough to start a conversation with our engineers about which layers your team can own at cut-over, which need a backstop and for how long.