And considering that the average garden-variety Pentium has several times the raw processing power of an IBM System/360 Model 195, one doesn't even have to shell out the big bucks for a System/z to get useful work done.
Windows may not like console applications... but Linux is quite happy to be used like plain old UNIX before they had the X Window System.
Well, it may take some efforts to understand the difference between mainframe and Unix server, and why its starting price is million dollars. The design philosophy is completely different:
- mainframe: nothing will fail, guaranteed. On Z9 every single command is executed on couple CPUs at the same time to make sure there is no CPU failure. Redundancy is built-in at the hardware level, every level down to memory, bus, I/O.
- Linux/Windows server: let's use the cheapest hardware and reply on software to handle failure. They are nothing but an up-scaled home PC. All the stuff about RAID, dual PSU, redundant network are all AFTER-THOUGHTS, and they don't guarantee anything.
Like I mentioned earlier, mainframe doesn't fit in Google/eBay's model, but it is just the best business machine. Imagine writing a banking transaction in COBOL, you just need couple lines and it works. If you do it on Linux systems, well, you just need to prepare to recover from every step, since every single step is possible to fail (what if disk failed at this given instruction, what if power outage, what if memory CRC error, a capacitor blows up on the motherboard?) I guess nobody cared about that kind of scenarios in your day-to-day life, but mainframe cares, and dealt with them already, 40 years ago. No hardware failure should ever impact the data consistency, period.
Also mainframe has huge I/O ability, the I/O sub-system is completely its own computers, many of them. Thus, all jobs are massively paralleled. Unlike on Linux server, a freaking disk read will take CPU interrupts. That's why you need the latest CPUs on linux servers. The best Linux server may be able to handle couple hundred threads at the same time (my data is not accurate), but back in 70s, SABRE system, every single airline ticketing window console is "directly" connected to the "one" mainframe and processing tickets at the same time.