Virtualization is a technology that uses software to divide one physical computer into several isolated virtual computers, called virtual machines. Each virtual machine runs its own operating system and applications as if it had the hardware to itself, while a thin software layer called a hypervisor shares out the real processors, memory, storage and network connections. Virtualization is the foundation of most private clouds, most public cloud compute services and many virtual desktop and network services.
At a glance
- One physical server can run many virtual machines (VMs), each isolated from the others.
- A hypervisor sits between the hardware and the virtual machines and allocates resources.
- It raises hardware use, cuts server counts, and makes moving, copying and restoring systems much easier.
- The technology also applies to desktops, storage, networks and network functions.
- Licensing, support and skills for the virtualization platform are now a major cost and risk item for many buyers.
What problem it solves
Before virtualization, the norm was one application per physical server, partly for stability and partly because operating systems and applications did not share a machine well. The result was rooms full of servers running at a small fraction of their capacity, each needing power, cooling, maintenance and eventual replacement. Building a new server meant ordering hardware and waiting.
Virtualization consolidates those workloads onto fewer, larger hosts. A new server becomes a few clicks instead of a purchase order. Because a virtual machine is a set of files, it can be copied, snapshotted, backed up as a unit, moved between hosts while running (on most enterprise platforms), and restarted on another host if hardware fails. Those properties also made disaster recovery, test environments and later cloud computing practical.
How it works
The hypervisor. A hypervisor is installed either directly on the server hardware (often called Type 1 or bare-metal hypervisors, the norm in data centers) or on top of a regular operating system (Type 2, common on laptops and desktops). It presents each virtual machine with virtual processors, memory, disks and network cards, and maps them to the real hardware.
Hardware support. Modern server processors include features that help hypervisors run virtual machines efficiently and keep them isolated, which is why the performance overhead is usually small.
Clusters and management. In production, several hosts are pooled into a cluster with shared or distributed storage. Management software lets administrators create virtual machines from templates, balance load across hosts, move running machines for maintenance and restart them elsewhere after a failure. Hyper-converged infrastructure packages compute, storage and virtualization into one clustered system.
Beyond servers. Virtual desktop infrastructure (VDI) runs user desktops as virtual machines in a data center or cloud. Network functions virtualization (NFV) runs routers, firewalls and similar functions as software. Storage virtualization pools disks from many devices into one managed resource.
In the cloud. Public cloud providers run their own virtualization platforms; customers rent virtual machines through Infrastructure as a Service (IaaS) without seeing or managing the hypervisor.
If you are weighing a refreshed on-premises platform against hosted options, our private cloud solution page explains the main choices.
When it matters for buyers
- When virtualization licensing changes. Changes in packaging and pricing from a major virtualization vendor have pushed many organizations to review alternative hypervisors, hyper-converged platforms, private cloud services and public cloud. Model migration effort and staff skills, not just license cost.
- When you inherit an environment. A new IT leader often finds aging hosts, unsupported versions, sprawl of forgotten virtual machines and unclear backup coverage. An inventory is the first step.
- At hardware refresh. Virtualized workloads are easier to move to new hardware, a hosted private cloud or public cloud than physical servers are.
- When planning disaster recovery. Replicating virtual machines to a second site or a recovery service is usually simpler than recovering physical servers.
- When considering bare metal or containers. Some workloads do better without a hypervisor, or with a lighter packaging model.
Questions to ask vendors
- Which hypervisor does your platform use, and how is it licensed (per core, per host, per virtual machine or bundled)?
- What tools do you provide to migrate virtual machines from our current platform, and what does a typical migration involve?
- How do you handle host failure, and how quickly are virtual machines restarted elsewhere?
- How are backup, replication and snapshots handled, and which are included in the price?
- What overcommitment ratios do you use for processors and memory on shared hosts?
- Which versions of our operating systems and applications are supported on your platform by their own vendors?
- What happens to pricing at renewal, and what notice do you give for licensing changes?
How it differs from containers
Virtualization and containers both let many workloads share one physical server, but they isolate at different levels. A virtual machine includes a complete guest operating system, so it can run different operating systems side by side and is strongly separated from its neighbors. A container shares the host’s operating system kernel and packages only the application and its libraries, so it starts faster and uses less memory, but isolation depends more on the host. Container management platforms such as Kubernetes orchestrate containers across many hosts, and in practice those hosts are often virtual machines themselves.