A state-of-the-art computer for mission critical tasks. Today, mainframe refers to a class of ultra-reliable servers from IBM that is designed for enterprise-class and carrier-class operations.
However, in the "ancient" mid-1960s, all computers were mainframes, since the term referred to the "main" CPU cabinet. The first mainframe vendors were (alphabetically) Burroughs, Control Data, GE, Honeywell, IBM, NCR, RCA and Univac, otherwise known as "IBM and the Seven Dwarfs." After GE and RCA's computer divisions were absorbed by Honeywell and Univac respectively, the mainframers were known as "IBM and the BUNCH."
IBM Is the Mainframe Vendor
For decades, IBM was the dominant vendor in the mainframe business. Although many tried to compete by offering compatible machines, they no longer do (see IBM-compatible mainframe
). HP, Unisys, Sun and others make machines that compete with IBM mainframes in many industries but are mostly referred to as servers. In addition, non-IBM mainframe datacenters have hundreds and thousands of servers, whereas IBM mainframe datacenters have only a few machines.
There Is a Difference
One might wonder why mainframes cost hundreds of thousands of dollars when the raw gigahertz (GHz) rating of their CPUs may be only twice that of a PC costing 1,000 times less. Here's why:
Lots of Processors, Memory and Channels
Mainframes support symmetric multiprocessing (SMP) with several dozen central processors (CPU chips) in one system. They are highly scalable. CPUs can be added to a system, and systems can be added in clusters. Built with multiple ports into high-speed caches and main memory, a mainframe can address thousands of gigabytes of RAM. They connect to high-speed storage subsystems that can hold petabytes of data.
A mainframe provides exceptional throughput by offloading its input/output processing to a peripheral channel, which is a computer itself. Mainframes can support hundreds of channels, and additional processors may act as I/O traffic cops that handle exceptions (channel busy, channel failure, etc.).
All these subsystems handle the transaction overhead, freeing the CPU to do real "data processing" such as computing balances in customer records and subtracting amounts from inventories, the purpose of the computer in the first place.
Mainframe operating systems are generally rock solid because a lot of circuitry is designed to detect and correct errors. Every subsystem may be continuously monitored for potential failure, in some cases even triggering a list of parts to be replaced at the next scheduled maintenance. As a result, mainframes are incredibly reliable with mean time between failure (MTBF) up to 20 years!
Here to Stay
Once upon a time, mainframes meant "complicated" and required the most programming and operations expertise. Today, networks of desktop clients and servers are just as complex. Large enterprises have their hands full supporting thousands of PCs along with Windows, Unix and Linux and likely some Macs as well.
With trillions of dollars worth of IBM mainframe applications in place, mainframes may hang around for quite a while. Some even predict they are the wave of the future!
Mainframes provided the computing power for major corporations for more than 50 years. Sperry Rand (Univac), IBM, GE, RCA, NCR, Burroughs, Honeywell and Control Data were the first companies to make mainframes in the U.S. This picture was taken in the mid-1970s. (Image courtesy of Unisys Corporation.)