Virtualization: hypervisors and virtual machines

·

Before virtualization, every critical application lived on its own physical server that rarely exceeded 10% CPU usage. The rest was expensive hardware consuming electricity and taking up space in the data centre. The hypervisor changed that equation by letting a single physical server run dozens of virtual machines isolated from one another. This article explains what a hypervisor is exactly, how type 1 and type 2 architectures differ, how VMware, KVM, Hyper-V and Proxmox compare, and which design decisions make the difference between a successful consolidation and a bottleneck.

What a hypervisor is and how it isolates workloads

A hypervisor is the software layer that abstracts the physical hardware and presents it to several virtual machines as if each had its own server. Its job is to distribute CPU, memory, storage and network among the guests while guaranteeing isolation: what happens in one virtual machine must not affect the others. That isolation is the fundamental security property of virtualization and relies on hardware extensions present in modern processors, which allow the guest's code to run in a controlled way without significant penalty.

It is worth distinguishing virtualization from containers. A virtual machine includes its own complete operating system and runs on top of the hypervisor; a container shares the host system's kernel and only isolates the user space. Virtual machines offer greater isolation and allow different operating systems to run on the same machine; containers are lighter and faster to start. They do not compete: in modern architectures they coexist, with containers running inside virtual machines.

Type 1 and type 2 hypervisors

Hypervisors are classified by where they run. Type 1, called native or bare metal, run directly on the hardware with no intermediate host operating system: VMware ESXi, Microsoft Hyper-V in server mode and the Linux KVM belong to this category. They are the data centre's choice because of their performance and smaller attack surface. Type 2, or hosted, run as an application on top of a conventional operating system — VMware Workstation, VirtualBox — and are convenient for development and testing on a workstation, but add a layer of overhead that makes them unsuitable for production.

Platform comparison: VMware, KVM, Hyper-V and Proxmox

The choice of platform shapes cost, ecosystem and vendor lock-in. VMware vSphere has been the enterprise standard for years, with a mature ecosystem for management, high availability and live migration, although recent changes to its licensing model have pushed many organisations to evaluate alternatives. KVM is the hypervisor built into the Linux kernel, open source, which underpins a large share of the world's public cloud; combined with management tools it offers enterprise-grade capabilities at no licensing cost. Hyper-V integrates naturally into Microsoft environments and is the logical choice where Windows Server already predominates. Proxmox VE, based on KVM and LXC containers, has gained traction as an open alternative with a complete management interface.

Comparison of virtualization platforms
PlatformTypeModelIdeal fitLive migration
VMware vSphere1 (ESXi)CommercialLarge enterprise, mature ecosystemvMotion
KVM1 (Linux kernel)Open sourceCloud, Linux infrastructureYes (with libvirt)
Hyper-V1Commercial / bundledMicrosoft environmentsLive Migration
Proxmox VE1 (KVM + LXC)Open sourceSMEs, labs, contained costYes

Consolidation, over-provisioning and its limits

Consolidation is the economic reason for virtualization: grouping many underused workloads onto fewer physical servers to raise hardware utilisation and cut electricity consumption and space. The key concept is over-provisioning, that is, assigning virtual machines more virtual resources than the host physically owns, betting that they will not all demand their peak at once. It works with CPU, because workloads rarely saturate simultaneously, but it is dangerous with memory: when physical memory runs out, the system resorts to swapping to disk and performance collapses. The practical rule is to be generous when over-provisioning CPU and conservative with RAM.

The other frequent limit is storage. Concentrating dozens of virtual machines on the same subsystem multiplies the input/output operations per second it must serve; a design that does not size storage by IOPS, and not only by capacity, ends up with latencies that affect all guests at once.

High availability and live migration

Virtualizing concentrates risk: if the physical server goes down, all its virtual machines go down. That is why production virtualization is designed in a cluster. High availability automatically restarts the machines of a failed node on another healthy node. Live migration — vMotion in VMware, Live Migration in Hyper-V — moves a virtual machine from one host to another without shutting it down, which makes it possible to maintain hardware or balance load without interrupting service. Both capabilities depend on shared storage accessible by all nodes and on a dedicated network with sufficient bandwidth.

Sizing the cluster's capacity properly means respecting the rule of reserving a margin node: if a four-node cluster operates at the limit of its aggregate capacity, the failure of one leaves the remaining three without the resources to take in the orphaned machines, and high availability fails just when it is needed most. Correct planning sizes the cluster to tolerate the loss of at least one node while keeping service running, which in practice means not sustaining a utilisation level above the threshold that keeps that buffer available.

Security and segmentation in virtualized environments

Concentrating workloads also concentrates the attack surface. Compromising the hypervisor is equivalent to compromising every machine it hosts, so the management plane must be isolated on its own network, accessible only through strong authentication and never exposed to the internet. Micro-segmentation brings traffic control down to the level of each virtual machine: instead of trusting that everything inside the data centre is safe, it defines what can communicate with what, so that a compromised guest cannot move laterally toward the rest. Information security management standards such as ISO/IEC 27001, and the guidance from INCIBE, offer a framework for governing these controls, the encryption of virtual disks and the patching of the hypervisor — a piece that is frequently neglected for fear of downtime and that live migration precisely allows to be updated without an outage.

Common mistakes and good practices

The most expensive mistake is over-provisioning memory the way you do with CPU, causing swapping to disk under load. The second is failing to separate management, storage and guest traffic onto different networks, which creates contention. The third is forgetting that virtual machine backups must be application-consistent: a snapshot taken in the middle of a database transaction can be unusable. The fourth common omission is not documenting the sizing or monitoring real utilisation, which makes it impossible to detect when the cluster is running short before users notice.

Frequently asked questions

Virtual machines or containers?

It is not an either/or choice. Virtual machines isolate better and allow different operating systems; containers are lighter and start in seconds. The usual practice is to run containers inside virtual machines, combining the isolation of one with the agility of the other.

How much can you over-provision CPU?

It depends on the workload profile, but ratios of several virtual processors per physical core are common in lightly loaded workloads. The key is to monitor the time the machines spend waiting for CPU: if that indicator grows, you have passed the limit.

Is it viable to move away from VMware to an open alternative?

More and more organisations are evaluating it after the licensing changes. KVM and Proxmox offer comparable capabilities for many cases, although the migration requires planning the disk conversion, training the team and validating high availability before moving production.

Does virtualization degrade performance?

With type 1 hypervisors and hardware extensions, the penalty is small for most workloads. Performance problems usually come from poor storage or memory sizing, not from the hypervisor itself.

Conclusion

Virtualization has gone from being a cost-saving technology to becoming the foundation on which practically all modern infrastructure is built, from the on-premises data centre to the public cloud. But consolidating is not stacking virtual machines until something breaks: it requires sizing storage by IOPS, being conservative with memory, separating the networks and designing high availability in a cluster from the start. The choice between VMware, KVM, Hyper-V or Proxmox depends less on which is "best" and more on the existing ecosystem, the licensing budget and tolerance for vendor lock-in. At Summum Systems we design, size and migrate virtualization platforms with technical judgement and a continuity plan tested before touching production.