Slashdot Mirror
Linux Possibly Ported to IBM Mainframes
Jah-Wren Ryel writes "
Vnunet is reporting
that IBM has a version of Linux ported to their S/390 mainframe
architecture waiting in the wings. Apparently there are two versions, one that runs under VM (a kind of meta-os, sort of like VMware) and one that runs on the bare hardware." An "anonymous source" and "speculation from analysts" story. Nothing official from IBM. Please read and judge accordingly.
Still, 5 years ago they were talking about how great it would be to have one OS across the board so that you wouldn't have to retrain employees as you scaled systems up from PC's. At the time, they were talking about doing that with OS/2 and we all know where that went, but Linux is out of their control so will continue to gain popularity no matter what they do with it.
It'd be cool to be able to telnet to bldvmb and get a Linux session when I log in, I hope they do this :-)
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
...take it from someone who's discussed it with the guy who actually started it. Linux/390 isn't very special, really. When I last spoke with him (I'm sorry, I've forgotten his name.. I forget names easily. :P) it just ran on the bare hardware, and didn't support much at all. If anything.
Now, some info on the S/390 from someone who's not gotten to play with one, really, but has seen one.
The S/390 is a 'massively parallel' computer. Meaning that everything is parallel. The S/390 is capable of running just about any OS you throw at it, from Linux on an x86 host controller to Windows NT on that same host controller. You can run AIX on an RS/6000 host controller. Or OS/400 on the AS/400 host controller. It's designed to do massive processing while serving up literally thousands of hosts. Usually 'dumb' terminals over twinax (twin-prong coaxial), triax (three-prong coaxial), or RS232/RS242.
No, it's not meant to run Linux instead of OS/390. I wouldn't dare to say that it should, because in truth, it shouldn't. The S/390 is not a 'convenience' machine or a 'play' machine. It is a mainframe, and it needs a mainframe OS.
However, I see absolutely no reason why Linux shouldn't run on the S/390. Bear in mind; running on the S/390 does not mean replacing OS/390. It means SUPPLEMENTING OS/390. Say you have an S/390 handling most of your financial transactions, but accounting wants a website to keep track of it. Running Linux on an x86 host controller on an S/390 is the perfect solution. But say accounting wants to cut some major expenditures out of the budget; eliminating OS/390 isn't a good idea. Plain and simple.
OS/390 is a *VERY* mature OS, pretty much dating back to OS/360 (the similarities between OS/390 and OS/370 are very obvious) and as a direct result, is rock solid stable, extremely secure, and inherently reliable. Add that in to hardware that is designed to have decades of uptime. Add in the power to get the job done and then some. That's what the S/390 is about. It's a big-bucks big-iron machine meant to be your network-edge solution for ERP and transactions and whatever else you want to throw at it. It's not your webserver, it's not your fileserver. It's a mainframe.
However, I've noticed quite a few people are moving away from S/390 to the actually more powerful RS/6000's, which lack some of the features of the S/390. Okay, MOST of the features people look for in the S/390. Some RS/6000 models border on the commodity machine definition. Linux doesn't belong there, either. Yes, that's right, you're hearing it from someone who spends about 99% of his spare time working on porting Linux more thoroughly to the RS/6000. Linux doesn't replace AIX. Period. AIX is a mature OS, probably 7 or 8 years Linux's elder. AIX has a very stable and regular release and development cycle, and is built on principles that have been proven a million times over. It's inherently reliable, stable, and very fast. Unlike Linux, AIX does not just have 'general' releases for all RS/6000's with all architecture support. There is AIX for the RS/6000 F40 (Dual PowerPC 604e) and there is AIX for the RS/6000 Power260 (single POWER3). You can't mix and match those two or components from them. AIX is optimized at the hardware level extensively. Unlike my work, it's built on native platform, optimized on that platform, and meant for that platform.
Yes, every piece of AIX has a common code base. The compilers do the work. That's why it's built on the native platform. You can get AIX C/C++ compilers for PowerPC 604e, POWER2, POWER3, and so on. And they're designed to optimize and compile reliably. ANd they do it well. Better than gcc or egcs could ever hope to.
Linux/390 is a great project. Like I said; there's no reason whatsoever that Linux should NOT be able to run on the S/390. There's no reason Linux should not be able to run on ANY system. The question is, though, do you want to replace what that system is MEANT to run with Linux?
Not yet. Linux is still a long way off from being ready to do that. But maybe someday it will be ready.
-RISCy Business
your company here.
shelby != ford
Then go and download a precompiled kernel and bootloader from http://www.linas.org/linux/i370.html. Note that this is a port that Linas Vepstas and others have been working on for some time now -- it's different to the rumoured "official" Linux port by IBM to which this story refers.
"The invisible and the non-existent look very much alike." -- Delos B. McKown
The purpose won't fall within the realm of most users, but it would be to run Linux along with other legacy apps on the same machine. Lest anyone think this is an entirely stupid idea, the place I used to work had an S/390 mainframe running legacy apps that are unlikely to be changed anytime soon, and we wanted to make use of some of the OMVS features on the machine. OMVS is supposed to be a UNIX-compatible OS running on the S/390. The machine was perfectly capable, if incredibly overpowered and overpriced for the simple purpose of getting at the unused disk storage we had for it via UNIX-type software. But OMVS was a real pain. The reason? It just wasn't convenient to compile for. We could download software all day, but if we wanted to run something of our own design or just compile the latest version of some kind of freeware for it we were out of luck without an army of IBM technicians to help us out. I personally wasn't a real supporter of keeping the mainframe around, but Linux as a VM partition would have at least been infinitely preferable to the proprietary system they were using, in openness and plain convenience. There are lots of possibilities for a system like this running on the existing mainframes out there. It allows integration of UNIX software into a mainframe shop's army of old-school applications, and increased use of underutilized hardware.
Not to mention the specialized hardware IBM mainframes use. It's not just a CPU and a few busses you have to be worried about. Literally *everything* has its own controller. Code would need to be written for so many things besides simply the CPU that, when all is said and done, what you'd have would be a far cry from the Linux we all know and love.
To put it bluntly: I'll believe it when I see it.
- A.P.
--
"One World, one Web, one Program" - Microsoft promotional ad
"Remember when the U.S. had a drug problem, and then we declared a War On Drugs, and now you can't buy drugs anymore?"
i think this would be a excellent step for IBM, i do a lot of work regarding SAP, ORACLE, BAAN, etc. and IBM is probably the least represented hardware Ive seen, this should open the doors of the market for them by providing a somewhat standard API for developers to work against, instead of their proprietary architecture.
May the forces of evil be confused on the way to your inbox.
Cisco makes the ESCON channel adapter cards for hte 7000 series routers for exactly this purpose.
"if there is tuned software available for a platform, this is typically a better choise than 'agnostic' software such as linux".,
is valid, but I would counter that in many cases, the increased performance of hardware is making the cost of maintaining tuned software prohibitive.
Look at web serving, for example. Perhaps its simply the timliness of the web that is forcing IBM to market AIX boxes as webservers, but nonetheless its out there in print. In such a case I would argue that simply having two FreeBSD boxes instead of one AIX box is going to give you the equivalent aggregate performance.
Stock hardware and software are slowly closing in on the tuned, expensive alternative. AIX and Solaris are in genuine danger of being made obselete by Linux and FreeBSD, if hardware continues to get cheaper and faster simultaneously. As the number of applications for which Solaris or AIX is preferred continues to shrink, the cost of maintaining them will outpace their profitability.
There is this assummption that mainframes are ancient technology and are just dinosaurs waiting to die. I work for a company that is a large mainframe shop, and after checking out one of our 4 data centers, I was quite impressed with the technology involved in a modern mainframe system.
We have several processing boxes which are linked by fiber optics. They provide many different logical partitions (i.e. systems) which are dynamically allocated across the available CPU's. Kind of a cross between redundancy clusters and Beowolf clusters with dedicated I/O processors to handle the I/O requests of several hundred users.
While these systems might not beat out the raw horsepower of modern CPU's, the supporting communication and I/O bandwith cannot be beat.
Now let's have Mindcraft have a benchmark against a mainframe running Linux and the best NT system (I know...it's an oxymoron) out there, and let see what results they get...
VM supported TCP/IP daemons as virtual machines in the late 1980's. I know, I installed it then to support a network which included RS/6000 machines running AIX, IBM's version of UNIX, connecting to the mainframe via an 8232. On MVS, in the early 1990's, you had a choice of running TCP/IP in outboard controllers, or (somewhat later in the 1990's) running TCP/IP in address spaces on the mainframe itself. In the early 1990's, IBM's proprietary System Network Architecture evolved to support flowing TCP/IP over VTAM (Virtual Terminal Access Method, or something like that), and VTAM itself became a marketing item called "AnyNet" ...charming, no? AIX continues to have an SNA-compatible networking product, but the die has long been cast, TCP/IP is now and has been for a long time, the lingua franca of networking. I also worked with the Amdahl UTS product mentioned in these posts and it worked with 3270 controllers, in the same late 1980's timeframe. AIX/370 ran on IBM 370 mainframes in the late 1980's. AIX/ESA ran on 380 or 390 mainframes in the late 1980's. The MVS operating system, since MVS/ESA 5.3, now known as Open Edition/390 or whatever the current marketeering name for it is, since the mid 1990's has included the "omvs" command, which, issued from a TSO session, will bring up a POSIX shell. You can use the standard command line interface; your directories and files are stored in an MVS dataset called a PDS/E; even though you can store NFS-mountable ASCII data in these essentially EBCDIC operating system datasets, they can be backed up and managed via standard MVS/ESA job control language batch jobs. In other words, with the exception of specialized super computers like Asia Pacific - Blue (which is a large cluster of RS/6000 boxes, by the way), the largest UNIX file server you ever saw is probably an IBM mainframe. Of course, you may never have seen a mainframe. They are still kept inside the raised-flooring area, minded by systems programmer gurus who dwell in closed offices. No one talks to the gurus, we just shove food under their doors with our paltry requests for guidance. Or so the story goes. Although new mainframes are expensive to purchase or lease, the vast number of users they can support actually make the total cost of ownership lower than our favorite mini-computer boxes running proprietary operating systems. That's been documented, you could look it up. You could get an older, water-cooled mainframe for the cost of hauling it away. The maintenance costs on water-cooled mainframes, and the proprietary software charges, will kill you though. Maintenance costs on air-cooled mainframes are much lower, but the software is still expensive... not just the operating system, but the extremely high cost of products sold by vendors. I know an organization that got a freebie mainframe, set it up, and found to their amazement that they were spending almost a million dollars a year on software charges (most of it going to application vendors) and maintenance contracts (oops, it was a water-cooled unit). Enter Linux/390, and the opportunity to replace some proprietary software applications with Linux based solutions. Hey, you can run these in a virtual machine with your regular mix of second-level virtual machines on top of the VM/ESA Control Program, or if you are Open Edition/390 based already, you can partition off part of your RAM and run Linux/390 on the bare metal. That's a pretty attractive proposition.
As for your list of problems:
And if you consider the version running under VM (VM/Linux? Linux/VM? ye gods..)...
rms will probably ask you to call it VM/Gnu/Linux or Gnu/Linux/VM
;)
"The obvious mathematical breakthrough would be development of an easy way to factor large prime numbers." Bill Gates,
No, I'm not talking out of my ass. I worked at a shop that was an ideal IBM shop (running OS/390, CICS, etc.) and were trying to implement Java (E-commerce and EJBs). IBM kept giving us the run around on Java support and OS/390. Our feeling was that they don't want to support it. If they open the door to Java-land, then why use IBM servers? Unless of course, they beat the crap out of everyone else's Java implementation.
We eventually went with Sun's E10k servers. We tried, but IBM basically just said that Java was not meant for their high-end machines. In personal computing and maybe RS/6000 land, Java is not deemed as a threat, but as far as enterprise level computing goes, they are SCARED or clueless (your pick).
I also said that chances are that it will not replace OS/390. See the post from the guy who has talked to the Linux/390 developer. It'd be nice if they ported CICS and VTAM (and everything else) to Linux. But, it ain't going to happen. They have too much invested in them to let them walk to another OS. IBM may release Linux/390 so that in our case where we want to use our IBM hardware we can use *different* software that is not part of the OS/390 fold, but is supported by Linux (et al).
Later,
Justin
Mu. P.S. The address you see is real. =)
IBM is a big company. A very big company. A very big company in the computer business. Very big companies tend to have quite a lot of smart people working for them (just being big enough makes that happen - they also have lots of dumb people working for them). And because they are in the computer business, they have smart people with all kinds of computer related interests working for them.
I wouldn't be surprised this started off by some IBM engineers trying to port Linux to the mainframe, just for the heck of it. And from there, it trickled upwards.
And what IBM gets out of it. Publicity. The ability to run applications without needing to port it - not even to AIX. An extra sales point. The investment in porting Linux might have been low, so IBM doesn't need gain much to make it worthwhile. And perhaps they are just thinking We offer Linux for the mainframe, just because we can.
-- Abigail
Is that for the component that actually implements the virtual machines, or for the OSes that run on the virtual machines, e.g. CMS (was CMS ever capable of booting as a single-user OS on a "raw" S/3x0?), or the regular IBM OSes (or non-IBM OSes, as per the topic of this thread...) that can also run as guests in a virtual machine?
As VM implements a virtual, err, umm, S/3x0, complete with channel controllers, simulated mainframe-flavored disks, etc., and, I think, depends on features of S/3x0 to provide that emulation, it probably couldn't be ported at all easily.
However, VMware implements a similar type of "virtual machine" on NT or Linux on a PC. (The posts asking whether VM was like VMware were somewhat amusing, given that VM/3x0 and CP/360 antedated the 8086, much less VMware, by many years, as in "probably around 15 years, if not more".)
Nope. It's a 16-general-register CISC instruction set (dating back to the early 1960's). The Linux on the IBM ESA/390 Mainframe Architecture page has a link to the the ESA/390 Principles of Operation manual, which describes the instruction set.
Okay, while I was not the person who talked to IBM directly (gee, you went to my website, good for you!), my understanding was that he talked to every person under the sky about getting our system to work with Java. He talked to their marketing people who assured them that Java would be entirely supported by IBM. He then asked for an example (Alan Cox's "show me the code"). Their technical people failed to produce one line of code that would run under OS/390. Eventually after dragging several VPs on both ends into it, we finally got something running on the system (this was mid-summer). Mind you this was on their beta OS running on a test proc. slice - we needed this on our production environment by the end of the summer (a company that dependent on it is not going to run their production m/f on an beta OS - even IBM's). However, by now, I believe that IBM has officially released the OS version that supports Java natively (but I may be wrong - I am no longer at IM).
However, since we used CICS (as do most IBM shops), we wanted CICS/Java connectivity on OS/390. And, that was what ultimately killed it. While Java was technically supported by the OS, their killer app did not support it. What they said is that all of the COBOL would have to be rewritten to conform to Obj. COBOL standards, then the Obj. COBOL could call C++ wrappers which could then call Java (and that was only if we installed every beta they had). In something so performance driven, this was not an option (never mind beta code). We were trying to make this faster NOT slower. At the same time, we had to support the legacy COBOL code. Ingram has so much COBOL code that forcing a rewrite of any subset of the code becomes a logistical nightmare. We eventually settled upon having a J/Gate (Java->CICS) architecture. Far from our ideal, but at the time, it was our only option. IBM failed to deliver what their customer needed when they needed it. Now, in six or seven months, IBM may finally get their heads out of the sand and support Java in CICS in a reasonable manner. But, Ingram is now a lost cause in that respect.
So, to clarify my position a bit, yeah, OS/390 supports it. CICS doesn't. If you aren't using CICS, then why use OS/390? Yeah, DB2 and all of that is supported in OS/390, but IMNSHO CICS is still the lifeblood of the OS/390 series...
BTW, you are indeed correct, IBM's machines kick Sun's ass clear across the room. And, in a place where IBM has so much clout, they should never have let Sun in the door. Now that Sun has their foot in the door, some are seriously considering dumping all IBM products and turning into a strictly Sun shop.
Sun delivered, IBM did not. That is what matters in the end...
Later,
Justin
Mu. P.S. The address you see is real. =)
Err, umm, once upon a time, mainframes were one of the few sorts of "normal computers" around. They don't look like PC's, but PC's are the only type of "normal" computers if you take "normal" literally, as in "average", as in "the average computer, by sheer numbers, is probably a PC" (I neglect embedded systems here, which I suspect may well outnumber even "IBM-compatible PC's").
The S/3x0 instruction set is pretty conventional - 32-bit, 16 general registers, register-register/register-memory/memory-memory instructions, most of which are boring old load, store, add, subtract, multiply, divide, etc., with various more exotic add-ons. Just because something's a mainframe, that doesn't mean its instruction set and CPU are immensely exotic.... (The Burroughs mainframes, and their Unisys A-series successors, have a fairly exotic instruction set, but IBM mainframes don't.) The ESA/390 Principles of Operation manual documents the S/390 instruction set.
If by "the VM" you mean "VM/390" or whatever it's called these days:
Perhaps you're thinking of System/38 and AS/400, where the compilers used by application programmers don't generate native machine code, they generate code for a virtual machine, and the low-level OS code ("system licensed internal code") translates that code into the native machine code for the particular machine, if it hasn't already been done, in order to run it (that native machine code being a System/3x0-like instruction set on older machines, and an extended flavor of PowerPC on newer machines).
A port is in progress, according to the Linux on the IBM ESA/390 Mainframe Architecture page. ("A port of glibc has been started. System calls work. Signals don't.")
Perhaps porting the X server code would make no sense (although there do exist graphical terminals for mainframes - I think they're still used for engineering and scientific work), but the X client code might be useful.
As far as I know, "SMP", meaning "symmetrical multi-processing", as in "multiple processors, without particular processors being devoted to particular tasks such as 'one processor runs OS kernel code and another runs user-mode code' or 'only one of the processors is allowed to ever run kernel code' (as opposed to, say, a single kernel lock allowing only one processor at a time to run kernel code), has, as a term, been around longer than have SMP systems with Intel processors. SMP systems, whatever they've been called, have definitely been around longer than have SMP systems with Intel processors....
Indeed? If so, then those "Power or PowerPC cores" are presumably interpretively executing the System/3x0 instruction set, which is not, and has never been, a derivative of POWER. (POWER and its descendants are load/store 32-general-register RISC architectures; S/3x0 is a 16-general-register CISC architecture with register/memory arithmetic instructions.)
double-precision shift instructions,
Dude, what the f**k is a double-precision shift instruction??
I mean, do you just mean that you can shift up to 64bits? (that's what I would guess from the conetext).
--
"Subtle mind control? Why do all these HTML buttons say 'Submit' ?"
ReadThe ReflectionEngine, a cyberpunk style n
Little anecdote on S/390 processors. We run a 2003-205 S/390 box. Came with 1 processor, running 13 MIPS. We found it to be too slow, so we ordered an upgrade to a model 207 (24 mips). Cost 60K. The tech came in one night, put a floppy in the hardware console (it is an IBM pc, P-II/300), fiddled a bit, and viola! It was now a 207!
I queried this guy a bit on this black (blue?) magic. Seems the 200x series processors come with 6 CPU's in them. The clock oscillation rate of each, and the number that actively process are controlled by SOFTWARE! These things can be 6-way SMP, and over 140 MIPS by doing the same procedure... oh, yea, and forking over mega-bucks to big blue.
What an incredible business model! My (red) hat is off to the guys who came up with *that*.
That aside... I agree with another poster that Linux would be a damn sight better than the OE shell and Unix system Services. And yes, you can run the X server on the mainframe, and use your regular old X-client. But, again.,.. why?
Jim.
Wakko Warner wrote: Actually, there's kind of a fallacy operating here. Mainframes date from the era when there were only mainframes. They were expensive and big because computers were expensive and big. I remember the end of that era clearly; I was there.
Minicomputers followed. They were somewhat less expensive and newer technology. In theory, a very complete set of applications could have developed for minicomputers, and in fact quite a few applications did, but it became a marketing war, and the pitch was that somehow minicomputers were less capable than mainframes. It was an effective pitch but really there was a lot of marketing hype there.
Finally, as "mainframe" manufacturers began to develop a lot of the same newer technology to keep up with "minicomputer" manufacturers in terms of cost and size, the "personal computer" came along and threw the whole equation into a cocked hat. PC hardware certainly was initially less capable than mainframe hardware; the processor architecture featured such ancient constructs as an accumulator. Missing were things like orthogonal register sets; clean, simple instruction formats; end-to-end error correction.
But nothing said that PC hardware had to be less capable than mainframe hardware; it just didn't matter that it was, since it was inexpensive enough to be bought by multiple users in multiple departments. This allowed departmental users in the corporate world to bypass the huge project backlogs that most IS departments had developed and gave control of much more computing to those departments.
By the present day, it's clear that most computation is done on platforms that are considerably more powerful than the first PC's, but are descended from those platforms. It's also the case that more and more features originally found on mainframes are making their way to PC's, and this trend is likely to continue.
PC variants (not necessarily on Intel architectures) represent "where it's at" because anybody can buy them and stage them for particular applications. They're cheap and easy to operate. So they are certainly likely to accumulate a very rich feature set as time goes on.
The "mainframe versus personal computer" war has never been about hardware capabilities, per se. It's always been about who has control of computing, and clearly the PC won that war. At this point we can regard mainframes such as the S/390 as being relatives of PC's, in that they have to compete in the market to perform the same tasks that PC's routinely perform.
Given that the only "emulator" I saw mentioned in the article was for AIX, and that by AIX they probably mean AIX for RS/6000, and that most if not all current RS/6000's use the PowerPC member of the POWER family of architectures, and that Linux also supports PowerPC, that's possible.
However, what the article said was:
(emphasis mine). That suggests that there's no ABI compatibility involved, just API - Application Programming Interface - compatibility.
In any case, the only ways to provide the ability for IBM mainframes running OS/390 to run Linux binaries would be
IBM already have "a POSIX to OS/390 translation layer", in a sense - they have a UNIX-compatible environment, in the sense that it passed the UNIX 95 test suite, so at least some programs can presumably be recompiled to run in that environment...
...as long as they, say, don't assume that the characters "A" through "Z" or "a" through "z" are encoded as a contiguous set of values; their UNIX environment uses EBCDIC, not ASCII, as its character set. (Here's the home page for the OS/390 UNIX System Services.)
I infer from the article that part of the rationale for a Linux/390 port is to make it easier to port applications from UNIX environments - OS/390's UNIX environment may not be enough like "Real UNIX", implemented, as it is, atop a different OS, and using a different native character set, and so on, to allow quick porting, whereas Linux systems look enough like "Real UNIX" to me, at least, for me to consider them to be "Real UNIX", even though the Open Group don't yet have any Official Certification Results for any Linux system but do have one for OS/390.
If this is true (highly doubtful), this would definitely send shockwaves throughout the enterprise class server industry. If IBM believes Linux is ready to run on their heavy metal boxes, then some serious (re)consideration of Linux is going to occur in the next few months.
However, after personally seeing IBM run away from Java on the mainframes (running OS/390-MVS), I have to doubt this is true. IBM looked scared to death of Java on the mainframes. For the personal computers (i.e. with Jikes), they really seem to embrace Java, but on their enterprise class servers, they seem to be frightened to death of it. After all, if they support Java, then why not just use Sun boxes? Of course if they do such a thing, they'd have to do it better than anyone else (including Sun) - not to say that they aren't capable of this, but they'd have to try really hard. =)
This could also mean the beginning of the end of OS/390 (MVS) - maybe IBM finally decided that they no longer want to mess with having to recompile or support weird programs on their OS. Just give them a little VM (or actually processor slices most likely) and let them run their own little OS that will allow them to run their weird apps. Keep all of the VTAM and CICS stuff under OS/390 though. I'd be pleasantly shocked if they came out with full-blown support for Linux though... Oh, man, CICS Server on Linux/390 - oooh, wow - there would be a lot of people jumping into to learn Linux really quickly if that happened.
But, rumours are rumours for a reason. I'd be curious to know whether Linus knows about stuff like this - would a company tell him that they were porting Linux to XYZ hardware platform?
Later,
Justin
Mu. P.S. The address you see is real. =)
Of course you're right. By 'normal' I really meant 'commodity'. Mainframes are quite dissimilar from the desktops or workstations that most of us are used to.
Perhaps porting the X server code would make no sense (although there do exist graphical terminals for mainframes - I think they're still used for engineering and scientific work), but the X client code might be useful.
Indeed. But the only really interesting part about porting X would be the server side. The server side has to interface with graphics devices. I haven't looked at the XFree codebase, but I presume that there's very little platform specific code in the client.
As far as I know, "SMP", meaning "symmetrical multi-processing"
I believe that this was a term introduced by Intel, and I don't think that other multi-CPU architectures are described with this acronym, but I could very well be wrong. I also believe that the 390 architecture is massively parallel but not really symmetrical. As another poster mentioned that the 390 can have various CPU modules which might not even run the same instruction sets.
It is "UNIX" in the sense that it passed the UNIX 95 test suite, but it's not an AIX port - it's part of OS/390 and, as noted in another post, it's different from what you might think of as "real UNIX" in some ways; for example, it does not use ASCII as its character set.
Thus:
...that question might better be phrased as "Can anyone tell me what you gain from running a native UNIX-compatible operating system instead of Unix System Services?", in which case the answer may be (as per the VNUNET article) that a native UNIX-compatible operating system such as Linux may look "more like real UNIX" than even the it-passed-the-UNIX-95-suite Unix System Services in some ways - ways that might make it easier to port to Linux than to Unix System Services.
One might ask why they'd want to port Linux rather than, say, revive the old "real UNIX" port of native AIX they once had (I don't know whether they still offer it or not); I don't know whether it's because
Define "32 bit" and "64 bit". S/390's general-purpose registers are still 32 bits wide (it says "For some operations, two adjacent general registers are coupled, providing a 64-bit format", but, as I remember, that's been true since System/360, back in the early '60's, in that it had, I think, double-precision shift instructions, at least) and, whilst I think ESA/390 has some segmentation-like scheme to boost the address space size above 2^31, the instruction set still looks more 32-bit than 64-bit or whatever. The internal data paths of the implementation may be wider, but, if you go by internal data path widths or processor-to-storage data path widths, there are few if any 32-bit processors left....
Aside from the fact that BNA is a pain to interface to a cheap wintel box, TCP/IP makes it much easier to have a terminal emulator running on your windows box.
On a different note, many great GNU programs have been ported to these machines to make porting Java easier. Granted, getting any (what you might think of as normal) C program to run correctly on a Unisys A-series is a challenge. 48-bit words with signed-magnitude representations are entirely unexpected by a normal C programmer. Don't use shifts!
/ \
\ / ASCII ribbon campaign for peace
x
/ \
Possibly, but I seem to remember hearing the term before x86 MP systems were common (although they date back at least as far as the Sequent Symmetry, so they do go back a while).
Perhaps Digital^H^H^H^H^H^H^HCompaq don't say "SMP", but they sure say "symmetric multiprocessing" (admittedly, not "symmetrical", if one wants to be fussy) on the Digital^H^H^H^H^H^H^HTru64 UNIX home page.
There exist S/390 machines that have a lot of processors, but the Multiprise 3000 "enterprise servers" (every time I hear some marketoon say "enterprise", I wonder whether they intend to install the "enterprise" product in question on the bridge of NCC-1701) start out as uniprocessors and go up to big honking two-way systems.
I also have the impression that the MP S/390's are "really symmetrical", in the sense that there aren't particular S/390 processors dedicated to specific functions.
I suspect he's thinking of I/O processors, e.g. the processors that run the channel controllers (which I wouldn't be surprised to hear were PowerPCs these days), the communication controllers, etc. - the processors that run the applications, and the bulk of the OS, are S/390s, as far as I know.
I work in the mainframe world and although we run MVS, I have seen references to IBM running Unix on their S/390 line (maybe others). So, it shouldn't be too hard for them to get Linux to run too. Someone posted something about TCP/IP and seemed suprised it might be on a MF. Well even on our MVS box we use TCP/IP for everything. We even have SMTP and FTP servers running on it and it works great. I think it would be wonderful to have Linux as the base OS.
--Scott 8-}
As the article states, the issue is that the mainframe OS/390 API porting target gets addressed late, if even at all, by the developers. So what's wrong with the mainframe OS is probably two things. One, there isn't as big a market share as UNIX has. Two, just because it's different, the cost of porting applications becomes much higher. Linux/390 is basically the low cost route to putting a POSIX/UNIX API on the mainframe hardware, which still has advantages in performing massively parallel I/O to hundreds of devices. Have you ever worked on a computer that had over 800 disk drives attached through over 30 I/O channels?
While we may have no real interest in it for home and small/startup business purposes, IBM has a real business interest in positioning their mainframe hardware investment to large corporations and banks who are moving to the newer software systems like Notes (IBM owns Lotus) and SAP.
I see it as proof that Linux is a mature OS. So it happens to save a legacy 32-bit architecture for a few more years.
now we need to go OSS in diesel cars
..they're minicomputers. That's not just a nit-pick - programming for a mainframe, at least at the lower levels, is much nastier than on the relatively uniform minicomputer architecture. A mainframe feels closer to a tightly integrated network of special purpose devices than a single CPU system.
From a programming perspective, a VAX is much more like a souped up microcomputer than a mainframe. One reason why virtual machines are popular on mainframes is that they hide the really ugly parts of the system; not just from application programmers but from kernel programmers.
I can believe a Linux port to VM, but I'm much more skeptical about a port to the bare metal.
Caveat - my mainframe penance was on Unisys machines, so correct me if I'm off base about IBM's big iron. No such doubts about the minicomputers though.
This is Old News, in computer terms.
LinuxToday ran a story on this back in mid-October. In it, they referenced an article in the Danish version of ComputerWorld. The feedback comments to LinuxToday are interesting, and several of them pointed out one project's home page.
dragonhawk@iname.microsoft.com
I do not like Microsoft. Remove them from my email address.
were true why in the world would IBM replace AIX on their Big Iron? Linux is a nice OS that is finally getting some attention but maybe it's getting a little too much. It's at the point now where everyone has a Linux .plan so they can tout it for PR and then not really go anywhere with it. Porting Linux fully to their mainframes would require a huge rewrite of the kernel which would make it look alot like AIX. Whats the point of calling it Linux if you have to take out most of the kernel code to port it to the mainframe? PR dude.
I'm a loner Dottie, a Rebel.
What would the purpose of this be?
As the article says (you did read the article, right? *grin*), the main point would be to run Linux in parallel with other S/390 OSes like MVS. As everyone seems to be pointing out, Virtual Machines are very popular in the mainframe world, and it is quite common to run more then one OS at a time. Thus, Linux would be just one more OS.
The suggested application was Lotus Domino. I can also see web servers, application servers, general Internet servers, that sort of thing, being useful. Perhaps a company running a big back-end mainframe database would want to use Linux for the front-end interface, with (for example) Cold Fusion. I can see quite a few uses for it.
Is a big bank going to dump MVS and move to Linux on the S/390? No, of course not. That isn't the point.
Plus, there is hack value. We can now say with a fair amount of confidence that Linux is the most scalable OS on the planet. It runs on everything from large IBM mainframes to hand-held Palm Pilot devices.
I'll believe it when I see it.
It is already partly done, from what I understand. My comment here has links with details.
dragonhawk@iname.microsoft.com
I do not like Microsoft. Remove them from my email address.