Customer Zero: What we are learning moving our own cloud from VMware to Nutanix

Ian Yates, Solution Development Architect.

---

Most organisations running VMware are working out what to do next. Few of them need another reason to look at an alternative platform. What they want is to hear from someone who is actually making the move. 

Strategix is making it now, on the cloud its customers already run on: a multi-tenant platform, with production in Johannesburg and disaster recovery in Cape Town, carrying the live workloads of businesses that each sit in a tenant of their own. Every one of those environments moves, Strategix’s own included as one of the migration waves. This is not a lab, a pilot or a single company’s estate. It is a service provider moving a working cloud, with paying customers on it, while it keeps running.

“We are our own first customer,” says Steve Dalgarno, MD of Strategix. “Every service we will offer on the new platform, we are building on our own cloud first. We call it Customer Zero. You are not the pilot. We are.”

The migration is in flight, not finished, and the company is telling the story as it goes.

The move did not start with a technology review. Strategix was a VMware partner for more than 15 years, and that programme was the basis of its cloud business. Under Broadcom, the licence changed to every core in the cluster, whether customers used that capacity or not. Then Broadcom closed the VCSP programme to Strategix and to most of the providers in it.

“What we sold did not stop working,” says Dalgarno. “We simply stopped being allowed to sell it. It was a strategy change, not a judgment on anyone, and hundreds of providers got the same news. The hard part was not choosing a platform. It was accepting that we had to move at all, and that was taken out of our hands. Most organisations reading this still get to choose their timing.”

Strategix evaluated several alternatives, from an open source hypervisor to commercial hyperconverged platforms, with real numbers against each. Every option on the shortlist would run its workloads, which is why a feature comparison settled very little. What separated them came down to three questions: can we operate it, can we afford it once every cost is counted, and can we run it as a service for many customers at once.

The first question did the work. Each finalist was run end to end in a hands-on trial driven by Strategix’s own engineers rather than demonstrated by a vendor, and the option that looked best on paper did not survive it. The trials surfaced two everyday capabilities the team had relied on for years and never written down: rolling a machine back to a known good point in place, in a single action, and letting the service desk clear a failed routine task without an administrator at a command line. On Nutanix, both are ordinary actions in Prism, the same console the team uses for everything else.

“You won’t find either on a feature comparison,” says Dalgarno. “You find out about them on an ordinary working day, when you need them and they aren’t there.”

The other two questions were settled the same way. On cost, Strategix built the full cost model, running costs included, before choosing, and Nutanix’s service-provider licensing is metered per processor core, so the software bill grows with the platform rather than arriving upfront. On running it as a service, Nutanix’s multi-tenant tooling, including SP Central for service providers, builds each customer environment from a standard pattern, so adding a customer does not mean repeating the same manual work every time.

Before sizing anything, the team measured what was actually running, and the first finding was that no two sources agreed. Two counts of the machines in scope came out six apart, both looking right, because each drew the boundary in a slightly different place. Another figure was about 7% out for a reason as mundane as a unit label: one tool reports storage in binary units, another in decimal ones.

Sized on a year of real usage rather than on what each machine had been allocated, the estate needs about two-thirds fewer servers than a straight lift-and-shift would have bought. “Most VMware estates are allocated far more processors than they ever use,” says Ian Yates, Solution Development at Strategix. “If you size from allocation, you carry that waste straight onto the new platform and pay for it again.”

Licence and hardware is the half of the number every vendor puts on a slide. The half that decides whether a business case is true is the rest: people’s time, running two platforms side by side during the move, rebuilding disaster recovery, retraining and carrying the risk for as long as the migration runs. Strategix built its cost model with all of that in, before it chose a platform.

The order for the first block followed the same discipline. The team took the first proposed build apart, line by line, before asking for any discount, and cut the list price by 48%. More than 60% of that saving came from the drives. The first build specified mixed-use flash, rated for heavy daily rewriting; the estate’s measured write load does not need it, so larger, far cheaper read-intensive drives are enough. Nutanix keeps two copies of every block of data on different servers and spreads writes across every drive, so no single drive carries the load. The processor choice barely moved the number. Starting with a first block that grows as workloads move across cut the initial commitment by two-thirds and, because the software is metered per core, means paying for about 40% fewer cores at the start.

Design and build is where a VMware team finds out which habits do not transfer. Some of what Strategix has met so far:

• Some choices are one-way doors. Nutanix keeps the cluster’s own metadata on the first flash drives in each server, and drives added later are not used for it, so Strategix fitted five per server from day one rather than topping up later. The platform that runs the cloud itself is a separate small cluster from the one running customer workloads: easy to build that way, hard to split later. On VMware with shared storage, both could have changed afterwards. That is the honest trade-off: more of the design is fixed at purchase.
• Address ranges are planned upfront. Each customer environment is its own virtual private cloud in Nutanix Flow Virtual Networking, with a private address range from a Strategix managed pool, so customers can change how they connect later without re-addressing a single machine.
• Keeping addresses means keeping the networks. The migration tooling carries each machine’s address across unchanged, so the old networks must exist on the new platform for the whole move. It is one reason moving the data centre at the same time was never on the table.
• The ground moves under you. Part-way through the programme, Broadcom withdrew public access to a component the standard migration tooling relied on. Nutanix has responded with a new method in its Move migration tool that does not need it, and Strategix plans to use it, with a second route through its own backup platform held in reserve. The new method changed the network design.

“The hypervisor is the easy part,” says Yates. “The work is in everything the machine touches – backup jobs, firewall rules, monitoring, licences tied to hardware. The migration is a weekend. The periphery is the project.”

The platform decision is made, the estate is measured and reconciled, and the first block of the new platform is on order. The design and build standards are being written now, ahead of the hardware arriving. Next come acceptance testing and the first workloads going live, then the estate moves across in waves until the last workload is off VMware. Strategix is deliberately not putting dates on those stages in public: timelines move, and Customer Zero reports what has happened, not what is planned. The next instalment follows when the first workloads are live, including what went wrong.

With support for current VMware releases running out, most organisations will not get to choose their timing for long. For anyone starting now:

1. Start with your own data, and validate it twice. Not a vendor’s assessment of your estate – yours.
2. Cost the whole thing, not the licence. If the number came off a quote, it is not the number.
3. Decide who will run the platform on day one before you decide what to buy.

“Many organisations running VMware will make a version of the move we are making,” says Dalgarno. “We would rather show people the scars than sell them a brochure, so they can learn from our experience instead of finding it all out the hard way.”

Get a two-minute estimate of your own migration. For organisations assessing their next move, Strategix's estimator provides an indicative view of what a migration could look like based on your environment. Click here to start your estimate.

Book a consultation with Strategix. Speak to Strategix consultants about your environment, migration considerations and the lessons the company is applying in its own Customer Zero journey.