THE DEFINITION
Three tests, set in 1974
Popek and Goldberg called a virtual machine an efficient, isolated duplicate of the real machine: programs behave the same, slow down only a little, and the monitor controls the resources.
What gets copiedFROM THE ORIGINAL PAPERS
A copy of the
whole computer.
Every virtual machine is a clone of a computer that does not exist.
What gets copied
1974efficient, isolated duplicate
- Behave the same
- Run nearly as fast
- Stay in control
- Three tests for a machine clone
The idea is older than the PC. It was built in the 1960s, given a precise definition in 1974, and in 2000 researchers counted exactly which PC instructions stood in its way.
Updated
THE SHORT ANSWER
A virtual machine is
a duplicate, with rules.
The definition still used in textbooks
comes from a 1974 journal paper.
THE FIRST ONES
Built in Cambridge, Massachusetts
CP-40 and CMS became operational in 1966 on a modified System/360 Model 40. CP-67 followed, and VM/370 became available in 1972.
The early systems
THE PC PROBLEM
Seventeen awkward instructions
A 2000 study checked all of the roughly 250 Pentium instructions and found seventeen that broke a key rule, making the chip non-virtualizable in the strict sense.
See the numbersHOW MUCH IS COPIED?
Not every virtual thing
is a whole machine.
NIST describes several forms of virtualization, told apart by which layer of the computer they imitate. Pick one to see what it copies. The bar shows how much of the machine the guest sees as its own.
Descriptions follow NIST Special Publication 800-125 (January 2011). General information about how the technology is described, not security advice.
TWO NUMBERS WORTH KNOWING
A short definition,
a long list to check.
One number from the 1974 definition,
one from the 2000 check of a PC chip.
3 tests
Essential characteristics of a monitor
An essentially identical environment, at worst only minor decreases in speed, and complete control of system resources. The speed test rules out traditional emulators and complete software simulators.
THE DEFINITION · POPEK AND GOLDBERG, 1974
17 of ≈ 250
Pentium instructions that broke the rule
These sensitive but unprivileged instructions ran without raising a trap, so a monitor never got the chance to step in and simulate their effect.
THE PC CHECK · ROBIN AND IRVINE, 2000
The 1974 paper's main theorem says a monitor can be built if every sensitive instruction is also a privileged one. Robin and Irvine applied that idea to the Pentium in a USENIX Security paper.
FOUR WAYS TO READ THE HISTORY
Follow the idea,
not the brand names.
The original papers are short, freely readable and still the clearest account of what a virtual machine is meant to be.
Start with the 1974 definition
Read the first section of Popek and Goldberg: one page sets out the three characteristics in plain words.
Then read how it began
Creasy's 1981 paper tells how CP/CMS was conceived in 1964 and why the design split machine management from user support.
Look for the trap rule
The key idea is simple: an instruction that touches the real machine must hand control to the monitor first.
Match it to today's names
NIST's 2011 guide maps the idea onto bare metal and hosted hypervisors, containers and application virtual machines.
THE EARLY SYSTEMS
One machine,
many private copies.
R. J. Creasy wrote that CP/CMS was conceived in 1964 as a second-generation time-sharing system. CP gave each user a separate computing environment at the machine instruction level, and CMS supported the user inside it. It was first called a pseudo-machine time-sharing system.
In 1966 CP-40 and CMS became operational on a System/360 Model 40 fitted with an associative memory for address translation. CP-67 ran on the Model 67, and VM/370 arrived in 1972. By 1981 Creasy counted over 2500 systems in operation.
Questions
Questions people ask
Who defined the virtual machine?
The definition most often quoted comes from Gerald Popek and Robert Goldberg, in Communications of the ACM in July 1974: an efficient, isolated duplicate of the real machine.
Is an emulator a virtual machine?
Not in the 1974 sense. The efficiency test requires most instructions to run directly on the real processor, which rules out traditional emulators and complete software simulators. NIST today treats hardware emulation as a type of hosted virtualization.
What was the first virtual machine system?
R. J. Creasy wrote that CP-40 and CMS became operational in 1966 at IBM's Cambridge centre, on a specially modified System/360 Model 40. CP-67 followed on the Model 67.
Why were PCs hard to virtualize?
Robin and Irvine found in 2000 that seventeen of the roughly 250 Pentium instructions were sensitive but unprivileged, so they ran without letting the monitor step in.
What is the difference between bare metal and hosted?
NIST says a bare metal hypervisor runs directly on the hardware, while a hosted one runs on top of an ordinary operating system. Servers most often use the first and desktops the second.
Where the facts come from
- Popek and Goldberg (1974), Formal Requirements for Virtualizable Third Generation Architectures, Communications of the ACM 17(7) (course copy, Carnegie Mellon University)
- Creasy (1981), The Origin of the VM/370 Time-sharing System, IBM Journal of Research and Development 25(5) (course copy, UMass Amherst)
- Robin and Irvine (2000), Analysis of the Intel Pentium's Ability to Support a Secure Virtual Machine Monitor, USENIX Security Symposium
- NIST SP 800-125 (2011), Guide to Security for Full Virtualization Technologies
- SHARE: SHARE and the History of Mainframe Virtualization (2017)