Virtualization — Trivia & Interesting Facts¶
Surprising, historical, and little-known facts about virtualization.
IBM invented virtualization in the 1960s — it was forgotten and reinvented 40 years later¶
IBM's CP-40 (1967) and CP/CMS (1968) were the first systems to implement virtual machines, allowing multiple users to run independent operating system instances on a single mainframe. The technology was mature by the 1970s but was largely forgotten by the x86 world until VMware reinvented it for commodity hardware in 1999. When told about IBM's prior work, VMware's founders acknowledged they were solving the same problem on different hardware.
VMware was founded in 1998 and produced the first x86 virtualization product in 1999¶
VMware was founded by Diane Greene, Mendel Rosenblum, Scott Devine, Edward Wang, and Edouard Bugnion at Stanford University. Their key insight was that x86 hardware was "unvirtualizable" (certain privileged instructions didn't trap properly) but could be made to work through binary translation — dynamically rewriting guest OS instructions. VMware Workstation 1.0 shipped in 1999, and it changed computing forever.
The Popek-Goldberg theorem (1974) defined the requirements for virtualization¶
Gerald Popek and Robert Goldberg published formal requirements for virtualizable architectures in 1974: all sensitive instructions must be a subset of privileged instructions (so they trap to the hypervisor). The x86 architecture violated this requirement — certain instructions behaved differently in user mode vs. kernel mode without trapping. This is why x86 virtualization required either binary translation (VMware) or paravirtualization (Xen) until hardware support was added.
Intel VT-x and AMD-V hardware virtualization were added in 2005-2006¶
Intel added VT-x (Vanderpool) to processors in November 2005, and AMD added AMD-V (Pacifica) in May 2006. These hardware extensions finally made x86 properly virtualizable without binary translation tricks. Ironically, early hardware virtualization was slower than VMware's binary translation because the VM exit/entry overhead was high. It took several processor generations before hardware virtualization became faster than software techniques.
Xen was created as a PhD project at Cambridge and powered the early Amazon cloud¶
Xen, developed by Ian Pratt and Keir Fraser at the University of Cambridge as a research project, was released in 2003. Amazon Web Services used Xen as the hypervisor for EC2 from its 2006 launch until 2017, when they began migrating to their custom Nitro hypervisor (built on KVM). Xen's paravirtualization approach required modifying guest operating systems but offered better performance than full virtualization on pre-VT-x hardware.
KVM was merged into the Linux kernel in 2007 with remarkably little code¶
KVM (Kernel-based Virtual Machine), created by Avi Kivity at Qumranet, was merged into Linux 2.6.20 in February 2007. The initial patch was remarkably small — under 10,000 lines of code — because KVM leveraged the existing Linux kernel for scheduling, memory management, and device drivers. This "turn Linux into a hypervisor" approach was so elegant that Red Hat acquired Qumranet in 2008 for $107 million primarily for KVM.
Containers are NOT virtual machines, but the confusion persists¶
Containers (Docker, LXC) use Linux kernel features (namespaces, cgroups) to isolate processes without running a separate kernel. Virtual machines run a complete OS with its own kernel on emulated hardware. The security boundary is fundamentally different: a VM escape requires exploiting the hypervisor, while a container escape requires exploiting the kernel (a larger attack surface). This distinction matters enormously for security-sensitive workloads.
The "noisy neighbor" problem drove most of the innovation in cloud virtualization¶
When multiple VMs share physical hardware, one VM's workload can degrade another VM's performance through shared resource contention (CPU cache thrashing, memory bandwidth, disk I/O, network bandwidth). AWS's development of the Nitro system, which offloads network and storage I/O to dedicated hardware, was driven largely by the need to eliminate noisy neighbor effects and provide consistent performance.
Live migration of VMs was considered impossible, then became routine¶
Moving a running virtual machine from one physical host to another without any downtime was considered impractical in the early 2000s. VMware's VMotion (2003) proved it was possible by iteratively copying memory pages while the VM ran, then briefly pausing to copy the final dirty pages. The total downtime is typically under 1 second. Live migration is now so routine that it happens automatically for hardware maintenance at every major cloud provider.
VMware's market cap peaked at $67 billion before the Broadcom acquisition¶
VMware's stock market valuation peaked at approximately $67 billion in 2019. Broadcom acquired VMware for $69 billion in November 2023 — the largest semiconductor industry acquisition in history. Post-acquisition, Broadcom's aggressive licensing changes (eliminating perpetual licenses, bundling products into expensive subscriptions) alienated many customers and drove significant interest in open-source alternatives like Proxmox VE and OpenStack.
Firecracker microVMs combine the security of VMs with the speed of containers¶
AWS created Firecracker (open-sourced in 2018) to power Lambda and Fargate. Firecracker launches a microVM in approximately 125 milliseconds with as little as 5 MB of memory overhead. It achieves this by implementing a minimal virtual machine monitor (VMM) that provides only the devices a function needs: network, block storage, serial console, and a partial keyboard controller (for the shutdown button). No USB, no GPU, no legacy devices.