I would really enjoy getting the chance to hack on a mainframe. I have the idea that 40 (50?) years of continuously developed OS from OS/360->today must be something really amazing to use and to work with. But, last time I looked it was nearly impossible to get a login to one.

I made an observation recently. I do a bunch of contracting for big companies:

Working in a Windows environment -- you will be given Administrator access to anything you ask for.

Working in a Linux environment -- you will be given a normal account and be able to SU to root

Working in a UNIX (Solaris/AIX/HP-UX) environment -- you will be given a normal account and the option to SUDO certain commands

Working in a zOS environment -- you wont be granted a login or allowed within 50 feet of the precious server.

You're not missing much. I may be jaded however from my experiences "hacking" in school and at RandomBigCorp in COBOL and JCL (not to mention the joy of using the typical editors in those environments). You can use other languages like C but in a typical corporate environment you're not going to see a lot of code in anything other than COBOL.

Writing code in assembly language on a mainframe my first year in college was actually a lot of fun. YMMV.

I forgot to mention that there are emulators out there. I have never used one but here's one from a quick google search: http://en.wikipedia.org/wiki/Hercules_(emulator)

When I was in school, IBM occasionally had z/OS programming contests and would provide an environment for student competitors.

Hercules is cool -- I last goofed around with it almost 10 years ago, but it was complete enough then to run VM/370 and MVS/370. If you're interested, start here: http://www.cbttape.org/vm6.htm and http://www.bsp-gmbh.com/turnkey/

"Continuous development" on the mainframe? Hah!

My first job out of university in 1996 was doing COBOL programming on mainframes for the federal government of Canada, and my first project was to rewrite a program originally written in 1972 that had changed very little since then. I have little doubt that after almost 15 years my rewrite is still much the same.

Anyway, I quickly realized that the mainframe environment doesn't exactly foster technical innovation so I moved on and am quite happy to have not sat in front a "green screen" in over a decade.

Your experience is with an application running on a mainframe and is probably typical. The OP was referring to the OS itself which is continuously developed with fanatical dedication to backwards compatibility.

That "fanatical dedication to backwards compatibility" works even harder towards glacial development for the OS than for applications. Or do you think OS developers will ship a new, faster scheduler, tweak the memory allocator and the thread scheduler every month?

Plus, you won't get five nines reliability if you tweak the kernel every week.

Mainframes are built to never stop, to never give you a headache and to never, ever surprise you.

Linux on zSeries doesn't run on z/OS at all. It runs on z/VM, a hypervisor that can either run purely as software, or hook into the PR/SM partitioning hardware built in the processors.

Why isn't Linux on the mainframe taking off? Mostly because it costs $75,000 per processor I'm guessing.

Not including the rest of the mainframe. Plus support. and you can't scale it in realtime.

Mainframes are absolutely fabulous creatures capable of amazing feats. But the culture/pricing around them has not kept up with the times. I wish it would have because a small mainframe would be a good choice for many purposes.

> and you can't scale it in realtime.

What do you mean by this?

You can't resize the images? Sure you can.

You can't add more virtual servers quickly? Of course you can. I'm told by a mainframe-operator acquaintance that Linux on z/VM can spin up hundreds of instances per minute.

You can't get the mainframe to run faster? Tosh. IBM generally oversupply you with CPU capacity; if you need more you can ring up and they will reconfigure them remotely to add CPU power.

Other advantages: a zSeries mainframe is extremely reliable, IBM's service is expensive but extremely thorough and unlike other virtual server environments it enforces partitioning at a hardware level, and much more completely. No more mysterious performance variation due to bursty neighbours.

> virtualizing servers by factors in the hundreds

Yeesh. The way I've heard it, this is due to a peculiar approach in mainframe development. Programs tend to assume isolated use of dedicated hardware, so they have to resort to creating ludicrous numbers of virtual mainframes rather than letting the programs coexist and interact within a single kernel.

"Programs tend to assume isolated use of dedicated hardware, so they have to resort to creating ludicrous numbers of virtual mainframes rather than letting the programs coexist" ----> Sounds like Windows VMWare/HyperV deployment justifications. :)

There is a supposed higher security in zOS partitioning -- but I have no idea what EAL Levels really mean: http://en.wikipedia.org/wiki/Evaluation_Assurance_Level

Mainframes predate the concept of multiple programs running at the same time. Truly. z/OS can handle running multiple programs at the same time, but it's optimised for batch operations.

If you want a multiprocessing OS, you can just use Linux on z/VM.

If one had too much cash sitting around and wanted to play with a low-end mainframe, what would be the ballpark cost to get one? Half a million?

Yes, I'm aware that IBM wants you to lease mainframe hardware rather than actually buy it.

If you had a truck and a bunch of free time, say, $499?

http://cgi.ebay.com/IBM-S-390-G5-Parallel-Enterprise-Server-...

The interesting thing about mainframe hardware is that it's designed to last forever, so the "old" stuff (which really means "brand new" stuff in mainframe terms) is still interesting.

The smaller z10s were being sold for less than a $100k. Bear in mind that they have a lot of inbuilt redundancy.