VM (operating system)

The heart of the VM architecture is a control program or hypervisor called VM-CP (usually: CP; sometimes, ambiguously: VM). It runs on the physical hardware, and creates the virtual machine environment. VM-CP provides full virtualization of the physical machine – including all I/O and other privileged operations. It performs the system's resource-sharing, including device management, dispatching, virtual storage management, and other traditional operating system tasks. Each VM user is provided with a separate virtual machine having its own address space, virtual devices, etc., and which is capable of running any software that could be run on a stand-alone machine. A given VM mainframe typically runs hundreds or thousands of virtual machine instances. VM-CP began life as CP-370, a reimplementation of CP-67, itself a reimplementation of CP-40.

Running within each virtual machine is another, "guest" operating system. This might be:

• CMS ("Conversational Monitor System", renamed from the "Cambridge Monitor System" of CP/CMS). Its official name is VM-CMS (confusing, since VM is commonly called VM/CMS). Most virtual machines run CMS, a lightweight, single-user operating system. Its interactive environment is comparable to that of a single-user PC, including a file system, programming services, device access, and command-line processing. (While an earlier version of CMS was uncharitably described as "CP/M on a mainframe", the comparison is an anachronism; the author of CP/M, Gary Kildall, was an experienced CMS user.)
• A mainstream operating system. IBM's mainstream operating systems (i.e. the MVS or DOS/VSE families) can be loaded and run without modification. The VM hypervisor treats guest operating systems as application programs with exceptional privileges – it prevents them from using privileged instructions (those which would let applications take over the whole system or significant parts of it), but simulates privileged instructions on their behalf. Most mainframe operating systems terminate a normal application which tries to usurp the operating system's privileges.
• Another copy of VM. A "second level" instance of VM can be fully virtualized inside a virtual machine. This is how VM development and testing is done (a "second-level" VM can potentially implement a different virtualization of the hardware). This technique was used to develop S/370 software before S/370 hardware was available, and it has continued to play a role in new hardware development at IBM. The literature cites practical examples of virtualization five levels deep (see page 55 of VM and the VM Community). Levels of VM below the top are also treated as applications but with exceptional privileges.
• A copy of the mainframe version of AIX or Linux. In the mainframe environment, these operating systems often run under VM, and are handled like other guest operating systems. (They can also run as 'native' operating systems on the bare hardware.)
• A specialized VM subsystem. Several non-CMS systems run within VM-CP virtual machines, providing services to CMS users such as spooling, interprocess communications, and specialized device support. They operate "behind the scenes", extending the services available to CMS without adding to the VM-CP control program. By running in separate virtual machines, they receive the same security and reliability protections as other VM users. Examples include:
  • RSCS ("Remote Spooling and Communication Subsystem", aka VNET) – communication and information transfer facilities between virtual machines[2]
  • RACF ("Resource Access Control Facility") — a security system
  • GCS ("Group Control System"), which provides a limited simulation of the MVS API.
  • Shared File System (SFS), which organized shared files in a directory tree

At one time, CMS was capable of running on a bare machine, as a true operating system (though such a configuration would be unusual). It now runs only as a guest OS under VM. This is because CMS relies on a hypervisor interface to VM-CP, to perform file system operations and request other VM services. This paravirtualization interface:

• Provides a fast path to VM-CP, to avoid the overhead of full simulation.
• Was first developed as a performance improvement for CP/CMS release 2.1, an important early milestone in CP's efficiency.
• Uses a non-virtualized, model-dependent machine instruction as a signal between CMS and CP: DIAG ("diagnose").

IBM coined the term "hypervisor" for the 360/65[3] and later used it for the DIAG handler of CP-67.

The Diagnose instruction ('83'x—no mnemonic) is a privileged instruction originally intended by IBM to perform "built-in diagnostic functions, or other model-dependent functions."[4] Under VM/370 it was repurposed for "communication between a virtual machine and CP." The instruction contains two four-bit register numbers, called Rx and Ry, which can "contain operand storage addresses or return codes passed to the DIAGNOSE interface," and a two-byte code "that CP uses to determine what DIAGNOSE function to perform."[5] A few of the available diagnose functions are listed below.

The early history of VM is described in the articles CP/CMS and History of CP/CMS. VM/370 is a reimplementation of CP/CMS, and was made available in 1972 as part of IBM's "System/370 Advanced Function" announcement (which added virtual memory hardware and operating systems to the System/370 series). Early releases of VM through VM/370 Release 6 continued in open source through 1981, and today are considered to be in the public domain. This policy ended in 1977 with the chargeable VM/SE and VM/BSE upgrades and in 1980 with VM/System Product (VM/SP). However, IBM continued providing updates in source form for existing code for many years, although the upgrades to all but the free base required a license. As with CP-67, privileged instructions in a virtual machine cause a program interrupt, and CP simulated the behavior of the privileged instruction.

VM remained an important platform within IBM, used for operating system development and time-sharing use; but for customers it remained IBM's "other operating system". The OS and DOS families remained IBM's strategic products, and customers were not encouraged to run VM. Those that did formed close working relationships, continuing the community-support model of early CP/CMS users. In the meantime, the system struggled with political infighting within IBM over what resources should be available to the project, as compared with other IBM efforts. A basic "problem" with the system was seen at IBM's field sales level: VM/CMS demonstrably reduced the amount of hardware needed to support a given number of time-sharing users. IBM was, after all, in the business of selling computer systems.

Melinda Varian provides this fascinating quote, illustrating VM's unexpected success:

The marketing forecasts for VM/370 predicted that no more than one 168 would ever run VM during the entire life of the product. In fact, the first 168 delivered to a customer ran only CP and CMS. Ten years later, ten percent of the large processors being shipped from Poughkeepsie would be destined to run VM, as would a very substantial portion of the mid-range machines that were built in Endicott. Before fifteen years had passed, there would be more VM licenses than MVS licenses.[6]

A PC DOS version that runs CMS on the XT/370 (and later on the AT/370) is called VM/PC. VM/PC 1.1 was based on VM/SP release 3.

When IBM introduced System/370 Extended Architecture on the 3081, customers were faced with the need to run a production MVS/370 system while testing MVS/XA on the same machine. IBM's solution was VM/XA Migration Aid, which used the new Start Interpretive Execution (SIE) instruction to run the virtual machine. SIE automatically handled some privileged instructions and returned to CP for cases that it couldn't handle. The Processor Resource/System Manager (PR/SM) of the later 3090 also used SIE. There were several VM/XA products before it was eventually supplanted by VM/ESA and z/VM.

VM's role changed within IBM when hardware evolution led to significant changes in processor architecture. Backward compatibility remained a cornerstone of the IBM mainframe family, which still uses the basic instruction set introduced with the original System/360; but the need for efficient use of the 64-bit zSeries made the VM approach much more attractive. VM was also utilized in data centers converting from DOS/VSE to MVS and is useful when running mainframe AIX and Linux, platforms that were to become increasingly important. The current z/VM platform has finally achieved the recognition within IBM that VM users long felt it deserved. Some z/VM sites run thousands of simultaneous virtual machine users on a single system. z/VM was first released in October 2000[7] and remains in active use and development.

IBM and third parties have offered many applications and tools that run under VM. Examples include RAMIS, FOCUS, SPSS, NOMAD, DB2, REXX, RACF, and OfficeVision. Current VM offerings run the gamut of mainframe applications, including HTTP servers, database managers, analysis tools, engineering packages, and financial systems.

As of release 6, the VM/370 Control Program has a number of commands for General Users, concerned with defining and controlling the user's virtual machine. Lower-case portions of the command are optional[8]

In the early 1980s, the VM group within SHARE (the IBM user group) sought a mascot or logo for the community to adopt. This was in part a response to IBM's MVS users selecting the turkey as a mascot (hilariously chosen, according to legend, by the MVS Performance Group in the early days of MVS, when its performance was a sore topic). In 1983, the teddy bear became VM's de facto mascot at SHARE 60, when teddy bear stickers were attached to the nametags of "cuddlier oldtimers" to flag them for newcomers as "friendly if approached". The bears were a hit and soon appeared widely.[9] Bears were awarded to inductees of the "Order of the Knights of VM", individuals who made "useful contributions" to the community.[10][11]

• Time-sharing system evolution