What a mainframe actually is
A mainframe is a computer built for one job above all others: doing an enormous number of small, boring, business-critical transactions per second without ever being wrong and without ever stopping. It is not 'a big server'. It is a different set of engineering trade-offs.
The plain-English version
Imagine a bank that has to answer the question 'does this card have enough money on it?' fifty thousand times a second, correctly, every time, for thirty years, while people are also loading new software onto the same machine. That question is not hard. Answering it a hundred billion times without losing one is very hard. Mainframes are the machines built for that second problem.
Most computers are designed to be fast at *computing*. A mainframe is designed to be fast at *moving data in and out and being certain about it*. The technical way to say this is that mainframes are optimised for I/O throughput, reliability and integrity rather than raw processor speed.
Three things that make it different
- Work is offloaded from the processor. Channel and I/O hardware handles the movement of data to and from disk and network so the main processors stay free for actual business logic. On other platforms the CPU does much more of this work itself.
- Everything is doubled, and errors are expected. Processors check their own results; memory corrects its own errors; a failing part can often be replaced while the machine keeps running. The design assumes hardware will fail and plans for it, rather than treating failure as an emergency.
- Sharing is the point. A single machine deliberately runs many isolated workloads at once, with the operating system policing how much of the machine each one is allowed. This is what lets one box do the work of a room full of servers.
The word 'mainframe' in practice
When people at work say 'the mainframe' they usually mean one specific thing from this list, and you have to work out which from context:
| They say | They usually mean |
|---|---|
| 'It's on the mainframe' | The data or the application lives on the platform |
| 'The mainframe is down' | One LPAR, one subsystem, or one region — almost never the physical machine |
| 'A mainframe job' | A batch job, i.e. a unit of work described in JCL |
| 'The mainframe team' | Anyone from systems programmers to COBOL developers to operators |
Common mistakes
The current generation of machines is designed and manufactured now, with current process technology. The software estate on top of it is often decades old — that is a different statement, and the one people usually mean.
A mainframe processor's clock speed tells you very little about how much business work the machine does, because most of the work it is measured on is I/O-bound, not compute-bound.
One physical machine is normally carved into many logical partitions. 'The mainframe' that went down at your site was almost certainly one partition or one subsystem inside it.
What you will see at work
- Nobody will ever show you the physical machine. You will work entirely through a terminal session or a web interface, and the hardware may be in a different country.
- Capacity is measured in MIPS or MSUs and it is charged for. You will hear people talk about a job's cost in a way that never happens on cloud teams — because on this platform, CPU consumption maps directly to a licence bill.
- The words 'the box', 'the CEC' and 'the CPC' all mean the physical machine.
Key terms
Check your understanding.
Take this lesson's quiz and save your progress. Free.