Mainframe computer

Categories: Mainframe computers

Image:Honeywell-Bull DPS 7 Mainframe BWW March 1990.jpg
A 1990 Honeywell-Bull DPS 7 mainframe CPU

Mainframes (often colloquially referred to as big iron) are large and "expensive" computers used mainly by government institutions and large companies for mission critical applications, typically bulk data processing such as censuses, industry/consumer statistics, ERP, and financial transaction processing.

The term originated during the early 1970s with the introduction of smaller, less complex computers such as the DEC PDP-8 and PDP-11 series, which became known as minicomputers or just minis. The industry/users then coined the term "mainframe" to describe larger, earlier types (previously known simply as "computers").

Contents

[edit]

Description

Modern mainframe computers have abilities not so much defined by their performance capabilities as by their high-quality internal engineering and resulting proven reliability, "expensive" but high-quality technical support, top-notch security, and strict backward compatibility for older software. These machines can and do run successfully for years without interruption, with repairs taking place whilst they continue to run. Mainframe vendors offer such services as off-site redundancy — if a machine does break down, the vendor offers the option to run customers' applications on their own machines (often without users even noticing the change) whilst repairs go on.

Mainframes often support thousands of simultaneous users who gain access through "dumb" terminals or terminal emulation. Early mainframes either supported this timesharing mode or operated in batch mode where users had no direct access to the computing service, it solely providing back office functions. At this time mainframes were so called because of their very substantial size and requirements for specialized HVAC and electrical power. Nowadays mainframes support access via any user interface, including the Web. Blade servers rather than mainframes are now increasingly requiring "exotic" cooling technologies.

[edit]

Market context

Nearly all mainframes have the ability to run (or "host") multiple operating systems and thereby operate not as a single computer but as a number of "virtual machines." In this role, a single mainframe can replace dozens or even hundreds of smaller servers, reducing management and administrative costs while providing greatly improved scalability and reliability. The reliability is improved because of the hardware redundancy noted above, and the scalability is achieved because hardware resources can be reallocated among the "virtual machines" as needed out of total system capacity. Mainframes can add system capacity nondisruptively, instantly, and granularly — what IBM marketing, with justification, describes as "on demand computing." Modern mainframes, notably the zSeries and System z9 servers, offer at least three major levels of virtualization: logical partitions (LPARs, via PR/SM), virtual machines (primarily via z/VM), and the protected and virtual addressing and workload management of their operating systems, notably z/OS. Together, these virtualization technologies support business-friendly mixed workloads, including Linux and Java, in a highly reliable and efficient manner.

Whether the acquisition cost of a mainframe makes sense will depend on its return on investment (ROI). Mainframe ROI, like any other computing platform, is dependent on its ability to scale, support mixed workloads, reduce labor costs, deliver uninterrupted service for critical business applications, and several other risk-adjusted cost factors. Some argue that the modern mainframe is not cost-effective. Sun Microsystems, Hewlett-Packard, and Dell understandably take that view, and so do some independent analysts. However, in recent years the majority view is that mainframes are uniquely valuable (and cost-effective) for many businesses and governments. IBM's zSeries and System z9 revenues are growing, despite declining prices, suggesting that view is indeed widespread among technology buyers. Major independent analysts, such as Gartner, reinforce that assessment.

There's no question that mid-tier servers are improving. Logical partitioning, at least in basic form, is now found in many high-end Unix-based servers. However, all servers are improving, not just non-mainframe servers. For example, zSeries/z9 servers execute each instruction twice, compare results, and shift workloads "in flight" to functioning processors, including spares, without any impact to applications or users. This rare feature, also found in Tandem Computers, is known as lock-stepping, because both processors take their "steps" (i.e. instructions) together. Not all applications absolutely need the assured integrity that these systems provide, but many do, such as financial transaction processing.

Undoubtedly debates will continue about the mainframe's value — and, for that matter, about the value of other computing platforms, such as the PC. The debate began with the 1964 introduction of the IBM System/360 and has continued for over 40 years.

As of early 2005, IBM mainframes dominate the market at over 90% marketshare. Unisys still manufactures ClearPath mainframes, based on earlier Sperry and Burroughs product lines. Fujitsu is nominally still in the market, servicing machines based on the former Siemens and Amdahl lines, while Hitachi has left the mainframe business (except for the zSeries 800 jointly designed with IBM). Bull's DPS mainframes are still on the European market. Acquisition costs vary, but new IBM mainframes start "under $200,000" (zSeries 890 Model 110, U.S. 2005 reported price, excluding disk storage).

[edit]

History

Several manufacturers produced mainframe computers from the late 1950s through the 1970s. In those "glory days" it was "IBM and the Seven Dwarfs": Burroughs, Control Data Corporation, General Electric, Honeywell, NCR, RCA, and UNIVAC. IBM's dominance grew out of their development of the 360 series mainframes; this basic architecture has continued to evolve into their current zSeries/z9 mainframes which is arguably the only mainframe architecture still extant that dates from this early period. That said, while they can still run 24-bit System/360 code, the 64-bit zSeries and System z9 CMOS servers have almost nothing physically in common with the older systems. The larger of the latter IBM competitors were also often referred to as "The BUNCH" from their initials (Burroughs, UNIVAC, NCR, CDC, Honeywell).

Notable manufacturers outside the USA were Siemens and Telefunken in Germany and ICL in Britain.

Shrinking demand and tough competition caused a huge shakeout in the market in the early 80s—RCA sold out to UNIVAC and GE also left; Honeywell was bought out by Bull; UNIVAC (a division of Sperry) later merged with Burroughs to form Unisys Corporation in 1986 (dubbed "dinosaurs mating"). In 1991, AT&T briefly owned NCR.

For a period of time companies found that servers based on microcomputer designs could be deployed at a fraction of the acquisition cost and offer local users much greater control over their own systems. "Dumb terminals" used for interacting with mainframe systems were gradually replaced by personal computers. Consequently, demand plummeted and new mainframe installations were restricted mainly to financial services and government. For a while, there was a consensus among industry analysts that the mainframe was a dying market as mainframe platforms were increasingly replaced by personal computer networks.

[edit]

Market rebound

That trend started to turn around in the late 1990s as corporations found new uses for their mainframes, since they can offer web server performance similar to that of hundreds of smaller machines, but with much lower power and administration costs. The growth of e-business has also dramatically increased the number of backend transactions processed by tried-and-true mainframe software as well as the size and throughput of databases. As of late 2004, IBM's mainframe revenues are increasing even with price reductions, thanks to attractive TCOs.

In 2005, numerous press stories detailed the loss of sensitive personal information from careless data handling. In one example, CardSystems, a U.S. credit card processor, discovered that thousands (if not millions — the exact number is unknown) of its credit card customers' records had fallen into the hands of hackers, thanks to a computer worm installed on one of its Microsoft Windows servers. The CardSystems breach cost an estimated $2.75 billion (U.S.), although the exact figure is difficult to ascertain. Financial institutions have not experienced these security-related failures with their mainframes, so many organizations are reassessing their entire data handling practices, often focusing on data recentralization on secure systems. In another example, ComAir's airline crew scheduling system, installed on a non-mainframe server, failed during the busy Christmas travel season in 2004. ComAir's Board of Directors fired its CEO as a consequence. Analysts expect a "flight to quality" to boost mainframe business given continuing high profile IT failures. (Mainframes do enjoy a justified reputation for reliability and security.)

Replacement of older 31-bit systems, including Amdahl and Hitachi models, should also positively affect IBM's mainframe revenues, especially in 2006 as those older systems reach their end-of-life. (z/OS will only support 64-bit systems from early 2007 onward.)

Another factor currently increasing mainframe use is the development of the Linux operating system, which is capable of running on many mainframe systems, either directly or, more commonly, in virtual machines. (See Linux on zSeries.) Linux allows companies and governments to take advantage of the software and development expertise from the open source community while enjoying the low per-user costs and high reliability (and security) of mainframes.

[edit]

Mainframes vs. supercomputers

The distinction between supercomputers and mainframes is not a hard and fast one, but generally one can say that supercomputers focus on problems which are limited by calculation speed while mainframes focus on problems which are limited by Input/Output and reliability. As a consequence:

  • Because of the parallelism visible to the programmer, supercomputers are often quite complicated to program and require specialized, task-specific software. In contrast, mainframes hide parallelism from the programmer. (One side effect is that even older software can benefit from adding mainframe CPs.)
  • Supercomputers are optimized for complicated computations that take place largely in memory, while mainframes are optimized for simple computations involving huge amounts of external data accessed from databases ("mixed workload").
  • Supercomputers tend to cater to science and the military, while mainframes tend to target business and civilian government applications. Weather modeling, protein folding analysis, and digital film rendering are all tasks well suited to supercomputers. Credit card processing, bank account management, market trading, and social insurance processing are tasks well suited to mainframes. (Exception: Certain military applications require high security, a mainframe strength.)
  • Supercomputers often run tasks that can tolerate interruption (for example, global warming forecasts and academic research). Mainframes tend to run those functions that must run reliably, even for years of continuous service (for example, airline bookings or credit-card processing).
  • Supercomputers are often purpose-built for one or a very few specific institutional tasks. Mainframes typically handle a wide variety of important, everyday tasks.
  • Mainframes assiduously and thoroughly support older software (dating back to applications written in the mid-1960s, in IBM's case) alongside new software. Supercomputers tend not to have backward compatibility as a central design feature.
  • Mainframes tend to have numerous ancillary service processors assisting their main central processors (for cryptographic support, I/O handling, monitoring, memory handling, etc.) so that the actual "processor count" is much higher than would otherwise be obvious. Supercomputer design tends not to include as many service processors since they don't appreciably add to raw number crunching power.

There's also some blurring of the term "mainframe" with high-end PC and Unix servers. (Some PC and Unix server vendors occasionally refer to their systems as "mainframes" or "mainframe-like.") That blurring of the term is not widely accepted, with the market in general agreement that true mainframes (particularly IBM zSeries) are genuinely and demonstrably different.

[edit]

Statistics

It has been reported that:

  • 85% of all mainframe programs are written in the COBOL programming language
  • 7% are written in Assembler, C or C++
  • 5% are written in PL/I
  • 3% are written in Java and other languages

Java use is increasing rapidly as of late 2004, and these figures are likely significantly out-of-date. (See also zAAP, WebSphere, and Linux.) Also, mainframe COBOL has recently acquired numerous Web-oriented features, such as XML parsing, with PL/I following close behind.

  • Most mainframes (rumoured to be 90%) have IBM's CICS software installed. Other software staples include IMS, DB2, and MQ.
  • In the early 1990s the media and many business and computing analysts predicted the death of the mainframe. The predictions were disproved as many companies embraced the mainframe as offering an affordable means to handle their Internet business models.
  • The quality of service offered by mainframes mean they are the preferred technology for many business critical applications.
  • As of late 2004, IBM claimed over 200 new (21st century) mainframe customers—that is, customers that had never previously owned a mainframe.
[edit]

Speed and performance

The CPU speed of mainframes has historically been termed as MIPS (million instructions per second). MIPS have been used as an easy comparative rating of the speed and capacity of mainframes. The smallest zSeries IBM mainframes today run at about 26 MIPS (z890 Model 110) and range up to well over 20,000 MIPS (in one System z9-109 Model S54 mainframe).

However, MIPS is inherently misleading for at least two reasons:

  • Changes in processor architecture over the years mean that MIPS ratings have long ago lost all direct relevance to the actual number of instructions executed. The meaning of MIPS is now purely for historical rather than for technical reasons.
  • A single value rating like MIPS is inaccurate when applied to different types of computer processing work. Mainframe design emphasizes "balanced performance," with tremendous I/O capacity in particular for superior online database and transaction performance. [See benchmark (computing) for a brief discussion of the difficulties in benchmarking such systems.]

Officially, IBM has long published a set of LSPR (Large System Performance Reference) ratio tables for mainframes that takes into account different types of workloads.

An analogy can be easily seen in today's competitive Unix server environment, where more demanding customers look at machine performance based on specific type of workload benchmarks, e.g. SpecInt or TPC-C, rather than looking at pure CPU clockspeed. However, increasingly even these benchmarks have their own problems.

Unfortunately, it takes a fair amount of work (and maybe guesswork) for customers to determine what type of workload type they have and then apply only the LSPR values most relevant to them. Use of MIPS still persists today to the extent that IBM and other consultants continue to publish MIPS for general reference.

[edit]

External links

de:Großrechner es:Ordenador central fr:Ordinateur central it:Mainframe nl:Mainframe ja:汎用コンピュータ pt:Mainframe ru:Мейнфрейм sl:Veliki računalnik sv:Stordator zh:大型计算机