Will Big Blue mainframes run Windows?
An obscure mainframe software company called Mantissa Corporation bragged last summer on the IBM VM listserv - which is dedicated to virtual mainframe environments - that it was creating a product called z/VOS that would allow slices of a Windows operating system to run atop z/VM, the hypervisor-as-operating system for IBM …
...if this really fits IBM's business strategy. They do have a lot invested in Windows on their X series servers so I'm not sure they have anything to gain by running Windows on the mainframe unless Microsoft is willing to pony up a lot of partnership. Right now the IBM mainframe line is pretty much a layered SOA setup with Linux and evidently possibly Solaris running on peripheral servers which can be used to run WebSphere and other web apps while the central mainframe processors running z/OS are mainly focused on DB2 and similar backend high availability vertically scaled applications. Thus it's not clear to me that IBM has much to gain by running Windows on their mainframes. The can run it on Nehalem instead at much lower cost and WebSphere is horizontally scaled so having multiple four socket blades is not an issue there. It appears to me that IBM will probably keep Windows on X class blades though if there is evidence they're creating a new "reduced capability" processor like the ones they use for Linux on the mainframe then this would have more credence.
The technology isn't the problem. After all, using "just-in-time" compilation technology, Intel was able to develop a software solution for executing x86 code on an Itanium that was faster than the hardware ability to execute x86 code that the chips included. And there's the similar approach Apple took to executing PowerPC code on Intel Macs. So relatively efficient software simulation is now well-established enough that we don't need to doubt that this is possible.
Some years ago (when microsoft ran their business on IBM midrange) I was at a midrange user group meeting where a senior IBM bod was asked if there were plans to run windows on midrange. The reply was that no they were not. It had been attempted but the attempt had to be abandoned because windows was too unstable. If you had to have windows you had to run it on an attached PC and let the midrange handle the i/o (which was quicker than letting windows handle the disk i/o locally). If they are running windows under z/os or VM it could provide some interesting response time stats.
who sees the humor in the possibility of heaing some actually say "big blue screen's o death"?
I'm sorry... but it had to be said.
I had heard a rumor a *LONG* time ago 10 years(?) maybe. Only, when I had heard it, it had been done by IBM types not OEM. I heard it from a pretty good reliable source so I believed it. A few years ago I was discussing it with other M/F types and was roundly criticized for telling a false hood. I will be happy to report back to them it is now semi official.
When I heard IBM got a SUN(?) box to run on the mainframe that showed to me that IBM is quite serious. Now with IBM probably buying out SUN its really serious.
IBM doesn't want this. This will create more players in the mainframe market which is 100% theirs at the moment.
They have a monopoly, why would they want to risk that?
IBM's customers keep asking... sorry, let me rephrase that.
The managers who run IBM's customers keep asking whether they can move stuff off x86 boxes and consolidate them onto the mainframe. Go look at the LINUX-390 mailing list and you'll see that this happens with enough regularity to make it an interesting proposition.
Given that the big boxes' forte has always been throughput, I'm curious as to whether the CPUs can actually handle the load these days (notwithstanding the various advances that have been made in the architecture).
I wonder when they will support all those old proprietary word processing systems that were popular during the 1970s and 1980s? I mean is it really a good strategy to shift legacy Windows code to Mainframes?
Wouldn't that be rather like retrofitting a Porsche to be horse-drawn?
It's hardly something new to create a true virtual machine environment (abstracting even the instruction set and hardware architecture) in order to run software designed for an alien environment. Indeed for many years you have been able to download for free a mainframe emulator called Hercules and run it on your PC. The difficult bit is making it run efficiently. Apple used a technology called Rosetta from Transitive Inc. in order to transition from Power to Intel architecture (although not a full machine emulator). The same technology was used to support SPARC executables under Linux. Transitive Inc is now owned by IBM - the perfect vehicle for running alien instruction sets on mainframes.
Of course it is a much easier to justify porting from expensive hardware to a (relatively) inefficient software emulator when moving to cheap commodity processors. Running Windows on a mainframe is an expensive proposition if CPU or memory usage is substantial.
But if you could successfully virtualise a lot of small Windows machines, this would be attractive to mainframe shops which have to tolerate the proliferation of pizza boxes ticking over at 5%. If you are not already using an x86 VM product, then rather than pick and adopt one, you may as well use the mainframe. Given the very low CPU usage of a lot of x86 boxes, you can likely cram quite a few onto tin which is quite happy running at nearer 100%.
Why on earth would you want to run a bloaty cack operating system like Windoze on a great piece of hardware like a mainframe?
Perhaps a mainframe is the only thing they could find with a spec sufficient to run windoze 7 ;o)
Deadly combination, eh ?
Paris, coz she and I make a deadly combination too
- "her brains and my money".. err. no "her money and my looks".. sorry got that wrong too.
- One last time: "her looks, my brains (which by now you might've guessed are no better than her brains)".
Surely, Unisys ES7000s have run Windows, in server slices, for almost a decade? Ironically enough, IBM are among the bigest customers for the ES7000.
A lovely new set of wheels for your Ferrari.
Very shiny, and perfectly square.
Why would anyone want this?
As an academic or research exercise, fine, nothing like taxing the grey stuff a bit, but as a serious end user product it makes little sense. why have the expense of an IBM mainframe, just to calculate where the mouse pointer should be on however many user desktops?
Maybe, it is to show that TSO isn't so bad after all!
As with most virtualisation games, the mere "doing" isn't actually hard. (Run Linux in your slice, run VMware on your Linux, ...) Doing it fast enough to be useful is another matter. Then again, "fast enough" depends on who you are talking to and what *they* perceive as the alternatives.
So what's the target market for this idea? Windows only real strength is on the desktop, and so unless you plan on rolling out Terminal Services on a frankly *terrifying* scale, I don't see where the customers are going to come from.
Finally, something that'll run Crysis properly. Now, if they can just sort out full screen anti-aliasing on a 3270 terminal........
One more challenge to add to the list then:
Windows on z/OS on Hercules on Linux.
Or even:
Windows on z/OS on Hercules on Linux on z/VM on Hercules on Mac OS.
Can't see any problems with any of that apart from licensing!
Bizarre to say the least, why would anyone want a general purpose OS designed for a general purpose CPU to run on a OS and piece of hardware optimised for through-put and multiple user access. Sounds a but like trying to steer Bugatti Veyron by tying a bit of string to wheel hubs and expecting to be able to steer the thing.
An interesting exercise though, in terms of how the underlying hardware works, in trying to get a piece of crap… let me rephrase that… a piece of software that depends on the concept of pushing addresses on to a stack to operate on a piece of hardware that doesn't have a stack (only addressable memory). I can see the microshit advertising already, no more BSOD, only a S0C4 addressing exception and 350,000 lines of a formatted core dump, please feel free to make the system email that to Redmond.
Of course the interesting question then will be will windoze be able to support a 3270 terminal with a light pen.
You likely can't ring them because their PBX runs on Windows, (as a guest machine under VM of course), and has blue screened.
Keith
Of course, when Windows NT was new, it ran on PowerPC and Alpha (and I'm sure that there was probably some MIPS and Precision Architecture ports as well).
At one time, over 15 years ago, there was an IBM project codenamed Prism that, according to a briefing I attended, would provide processor neutral I/O system and backplane that would allow multiple heterogeneous processors to be plugged in, and share I/O resources. At the time, they were talking about mainframe stuff, System 38 (this was before or about the time of AS and OS/400), UNIX (AIX) and Intel running OS/2. Never heard any more about it.......
.....but it is interesting that when VM got hardcoded into the firmware of whatever mainframe it appeared in, it was called (at least for a short time) Prism. Co-incidence, maybe?
IBM are currently trying to make as much of their technology cross-architecture anyway. Thus the merged iSeries and pSeries Power line. And a lot of the DS storage range shares technology with midrange. HMC's and SVN's are microcoded xSeries boxes. It's all coming together. And it all comes in big black racks that you cannot tell apart without reading the model numbers.
Black helicopter, because I was subject to a non-disclosure agreement at the time, and I wonder whether that still applies.
Attended Marist College when IBM started working with to port Linux to their mainframes. The project was a succcess, not so much because of the "brilliant minds" on staff, but due to the fact that the student population that worked in the IT department was heavily involved in the Open Source community. The guys (and a few gals) were ususally running some highly customized version of Linux on their PCs and I have fond memories of watching out "Into to Linux" prof spend most of the class trying to stay one step ahead of the students in the class who were hacking his machine to bits. Just goes to show that, whe it comes to IT< real world expirience trumps a diploma any day...
The world's least reliable, least secure operating system running on the world's most reliable, most secure hardware platform.
It wasn't PRISM - it was PR/SM (Processor Resource/System Manager). In fact Amdahl got there first with MDF (Multiple Domain Facility), although PR/SM was much the better product when it came out.
Cheaper than Zseries. High-end Pseries frames are bulletproof enough.
Give ESX a single large virtual CPU and avoid multi-CPU licensing problems with Windoze.
Sign up, sign up for Blocks and Files, The Reg's weekly storage newsletter
