You're Gonna Need a Bigger Mouse for that Mainframe

You're Gonna Need a Bigger Mouse for that Mainframe

Summary: At the SHARE conference in Austin, Texas, the Birmingham, Alabama-based Mantissa Corporation unveiled z/VOS, a hypervisor that enables Windows desktops and servers to be virtualized on IBM mainframes.I don't generally like to blog about IBM technology on Tech Broiler because as an IBM employee, much of what I say could be considered a conflict of interest -- so I try to stick with topics that are general to the industry and have broad appeal.

SHARE:

At the SHARE conference in Austin, Texas, the Birmingham, Alabama-based Mantissa Corporation unveiled z/VOS, a hypervisor that enables Windows desktops and servers to be virtualized on IBM mainframes.

I don't generally like to blog about IBM technology on Tech Broiler because as an IBM employee, much of what I say could be considered a conflict of interest -- so I try to stick with topics that are general to the industry and have broad appeal. This time, however, as it concerns a 3rd-party vendor, I'm making an exception.

Click on the "Read the rest of this entry" link below for more.

I've known about the development of Mantissa's z/VOS for about seven months, although I learned of it first here on Tech Broiler, via a comment posted by Mantissa's CEO, Gary Dennis, on an article I wrote about x86 possibly losing its relevancy. In that article, I suggested that Microsoft port Windows to z/VM natively, running on IFL processors, much like Linux runs on the architecture today.

After all, Windows NT was ORIGNALLY designed to have a portable kernel with a hardware abstraction layer, it once ran on PowerPC, NEC MIPS and Intel i860 RISC processors, and of course Windows still runs on HP's/Intel's Itanium. So a native port to z/Architecture seemed like it would make more sense then emulating x86 machines on a very foreign systems architecture, initially.

What I didn't count on is a vendor actually re-implementing the x86 environment within an emulated container on the z-Series. That just seemed too crazy, because the thunking and instruction translation challenges involved to achieve good performance seemed immense.

But when there is a will, then there is a way. And apparently, Mantissa does have a way.

Granted, I've never seen this product demoed. I don't have any architectural whitepapers from Mantissa, and I can only surmise as to how the thing actually works. The company has been extremely tight-lipped about their product. I will say this, however -- if it does work, then the "Screen" I wrote about last week just became a big step closer to reality.

Obviously, Mantissa will have a lot of hurdles to overcome -- Mainframes are not small purchases, even with "Baby Z's" so not just everyone is going to see IBM mainframes as a viable x86 virtualization or server consolidation solution. The price/performance ratio of Mainframe-based Windows hosting versus bare-metal or x86 virtual infrastructure via VMWare/Xen/Hyper-V/KVM will really have to be calculated and put to the test.

Then there are the legal issues that Mantissa may have to deal with from companies like Intel who might feel threatened by what this might mean to their core business if very large companies start getting the idea that they can run thousands of Windows instances on one IBM mainframe alongside native z/Linux and z/OS apps instead of on 1000 clustered servers.

There is also the challenge of overcoming a ton of performance issues -- since the x86 architecture is being completely emulated, there's all that I/O translation and such that has to occur. Does it make sense from a cost and performance standpoint to consolidate your MS SQL Server infrastructure, Sharepoint or corporate Exchange or Active Directory on a z10 T-Rex? I dunno.

Disclaimer: The postings and opinions on this blog are my own and don’t necessarily represent IBM’s positions, strategies or opinions.

[poll id="14"]

Reblog this post [with Zemanta]

Topics: Operating Systems, Hardware, IBM, Processors, Servers, Software, Windows

About

Jason Perlow, Sr. Technology Editor at ZDNet, is a technologist with over two decades of experience integrating large heterogeneous multi-vendor computing environments in Fortune 500 companies. Jason is currently a Partner Technology Strategist with Microsoft Corp. His expressed views do not necessarily represent those of his employer.

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

25 comments
Log in or register to join the discussion
  • Wrong way.....

    Rather than virtualizing the x86 on the zos, wouldnt it be better to virtualize the zos on the x86?

    One of the milstones around the neck of the computing industry as a whole is the licence price IBM recieves from its mainframe software. I think it is typically 120% of corporate profit.
    ynot56
    • No

      z/OS and mainframe hardware does what it does because it is designed to do batch processing at enormous rates. Running emulated z/OS on x86 would not give you the architectural scalability that very large companies need. We're talking about systems that have to do transactions in the tens of millions per day, far beyond what could be jury-rigged with x86 clusters or even with "large" monolithic x86 systems like Unisys's ES7000. It's a completely different class of performance. For small scale software development so that smaller vendors can get in on the z/OS development action I can see a need for it, but not for deployment.

      As to mainframe software -- I think you are missing the point. The IFL processors that z/Linux and Mantissa's z/VOS runs under are not metered like the GP processors used for z/OS. The whole point of the IFL is to run open systems and open source software, not IBMs. If you are going to virtualize thousands upon thousands of systems, particularly Linux, there's no more cost effective platform than a mainframe. You just cant do what a Mainframe does with x86 architecture.
      jperlow
    • Virtual Z on x86

      Already done, at least for the precursor to the current hardware. You too can run OS/MVT on your desktop. Check out www.hercules-390.org
      MrBeck
  • RE: You're Gonna Need a Bigger Mouse for that Mainframe

    First to jperlow, "Mainframe Batch" geesh, a computers a computer. The mainframe has been doing OLTP longer than Windows existed.
    Next you want 5 9s, run a Mainframe, that was possible 20 years ago.

    So this makes perfect sense with one exception that no one has answered (I do know why, but nobody even discusses it).

    WHY DO WE EVEN NEED VIRTUALIZATION ??

    Why can't you run 100, or a thousand apps under one copy of Windows, why can't one windows server support 100,000 exchange users, why can't you run many SQL DBs on a single windows server.
    If the hardware has the horsepower, why do I need two copies of windows to run 2 different applications ??
    ...
    OH, and Unix/Linux don't really get a break either, even IBM runs multiple AIXs on it's big UNIX hardware.

    Mainframes, ONE OS.

    Apps can't hurt each other, apps can't hurt the OS, apps can't take over the machine and bring it to it's knees. True Multi-Tasking.
    And if you consider CICS you get that at what Windows would refer to as thread level.

    So think about it, Mr. Bill are u listening ??? Oh thats right Stevie !!!!!

    Note I have been in the industry for over 40 years, I have worked on mainframes(DOS to z/OS), Unix, Windows Servers (3.5 to Today) and PCs. I have done application programming, and systems programming on mainframes and Windows, and most other functions in IT at one time or another.

    Oh, while at it, another question.

    Isn't cloud computing a Mainframe ?

    If it walks like a duck .....
    knudson
    • 1000's of apps under Windows??

      A previous blogger commented
      [i]"Why can't you run 100, or a thousand apps under one copy of Windows, why can't one windows server support 100,000 exchange users, why can't you run many SQL DBs on a single windows server.
      If the hardware has the horsepower, why do I need two copies of windows to run 2 different applications ??"[/i]

      From my viewpoint, you need multiple copies of windows to be run, is that Windows (and may be others as well) is not a "true" multi-tasking OS.

      (Just one example of Windows not being multi-tasking is in the file system, especially the "mount" logic. This is easily demonstrated in Windows - try and mount/load (say) a CD with sector zero errors. Result: windows just stops.)

      Please do not interpret my Windows comments above as "Windows bashing" - I actually like it, BUT it could be dramatically improved - but then again Windows could arguably be hamstrung by the X86 hardware.

      Comments.
      koala1515
      • You're quite correct about the CD errors etc.

        Perhaps Jason could start a topic:

        Quite-Probable Issues That Shouldn't Affect ANY SOFTWARE.
        errors with removable media #1
        dependencies on missing devices #2
        illegal characters in filenames #3
        insertion of io-system filters/hooks without a proper audit-trail. #4
        Custard_over_2x_Pie
  • So where can we get real performance data on the Z series?

    Wading through lots of marketing BS. :(

    TOP OF THE RANGE ENTERPRISE
    CPUs 4.4GHz (upto a maximum of 64)
    1.5TB of memory
    Point-to-Point SMP

    I'm not 100% clued-up on that last detail, although I do understand that it's either the cpu-cpu or cpu-memory routing/bus connection.

    But there's no raw data on what these monoliths can actually deliver.

    I can see what you're driving at though. Swapping bytes to emulate an x86 must be expensive in cpu cycles lost. Yes, it can be done in hardware, on integer quantities, but I haven't heard of byte swapping FP quantities. But, anyway you cut it, there's an awful lot of performance *lost* due to the x86 having little-endian byte ordering, and most mainframes being big-endian.

    Ideally, you would keep the VM entirely in memory, which sort of rules out sharing cpu's. Doesn't it?

    IMHO, we should all wait to see how the x86 market develops. When virtualization is seen more positively, perhaps then, we can finally rid ourselves of the legacy of x86 running on bare metal.
    Custard_over_2x_Pie
  • RE: You're Gonna Need a Bigger Mouse for that Mainframe

    Supposed to have been reply to comment, had an issue here and it ended up reply to story.

    Perfomance numbers should be available on the IBM website. Compared to even the biggest Intel they will probably be mind boggling.
    The Zs are natively MPs, but you don't know it, it looks like one giant CPU. The OS manages it, you can easily run on a different processor for each slice (your turn at multi tasking).
    knudson
    • Not anywhere near as fast as you think

      Here's the "real world" from somebody who has to support apps on Win/Linux/UNIX as well as on mainframes.

      Client/Server apps are not well suited to the mainframe environment as they tend to hog 100% of an available CPU. Also, C++ applications do not compile efficiently in mainframe environments due to mainframe compiler limitations. You are doing good when you are running the same C++ application at 1/2 the speed you could achieve on "current" client/server hardware. This is why scientific apps and CPU intensive apps like video rendering run on x86 server farms, typically using tweaked Linux implementations.

      Mainframes are very, very good at running assembler apps, COBOL apps, and massive database manipulations. They are also good at tasks requiring huge I/O such as distributed banking applications.

      Best case in this solution is that you will probably end up being able to execute 1 virtual machine per physical mainframe CPU for any CPU intensive task. For non-CPU intensive tasks like web hosting, you are better to run a USS or Linux LPAR on your mainframe and not burden it with the additional overhead of emulation. Mainframes are no longer built out of exotic parts, they are 99% commodity hardware or they would be completely uneconomical to build or sell. 4.4GHZ of clock speed is still 4.4GHZ regardless of the environment around it.

      However, the best solution of all is to use Websphere to link a high-performance Win/Linux/UNIX server to the mainframe OS and offload the non-mainframe suited tasks to it. Web services allows the 2 operating systems to each do what they are best at while maintaining close coupling between the 2 environment. In effect, the web services connection allows the server to act as an extra "virtual" environment.
      devo@...
      • Wrong, wrong, wrong!!!

        "Mainframes are no longer built out of exotic parts, they are 99% commodity hardware or they would be completely uneconomical to build or sell. 4.4GHZ of clock speed is still 4.4GHZ regardless of the environment around it."

        This is absolutely incorrect! The IBM System z is nowhere near commodity hardware (otherwise it would be far cheaper than it is and IBM could save millions of the dollars it spends on R&D.)

        Clock speed is only one factor to consider when selecting the performance characteristics of a platform. The processor architecture is far more critical to how a workload will perform on a given platform. Amount of cache, processing units, etc. all have a bearing.

        No space here to go into it all, but if you're really interested in being able to talk with some authority check out http://www-304.ibm.com/jct01005c/university/scholars/products/zseries/SystemzArchitectureCourse.pdf
        lostark98
      • A computer is a Computer, the OS is the KEY

        Coding is coding, if it runs good on one processor, it should run good on another. If it doesn't it is either using some unique trick, or just badly written. OK, MFs don't do graphics well, but actually neither does Intel, it takes a separate graphics card and GPU to handle that.

        C and C++ are available for the mainframe.

        Finally, learn how a real OS works (I am a Windows supporter and user, but ). Under Z OSs no application can monoplize the machine, the OS WILL time slice wether the code gives it up or not. ie TRUE multi-tasking, not cooperative as implemented in Windows AND Linux.

        Many of your comment do point to one of the main reasons we need VM:

        POOR CODING (application and OS).
        knudson
        • Night and day

          I agree with most of what you've said, but I disagree with the subject! The design is critical in determining the best platform for a given workload. The System z and z/OS are designed as a system from the chip up to do many things at once with high integrity and reliability and availability.

          The Intel x86 processor is designed to do one thing fast. Various manufacturers then architect their offerings by lashing multiple processors together in various ways to achieve tremendous performance. (Some better than others.)

          lostark98
  • RE: You're Gonna Need a Bigger Mouse for that Mainframe

    I can see a couple of problems here.

    1) all your eggs are in one basket. One hardware issue and the server with all of it's virtualizations are down. Yes you can shift them to another server, but how much money are you willing to spend?

    2) IBM servers require IBM hardware. And that stuff is just plain expensive. We bought $6k servers that had $10k mother boards and two $1k power supplies. Hard drives were about 150% of the price for the best server drives available at the time.

    3) One vendor means that if you don't get a problem solved to your satisfaction means that you have no other recourse. I have gotten the "It's a known problem, don't do that." line more than once.

    4) IBM used to be really, really big in the IT industry. They aren't quite so big today. Do I really need to remind you why people left?

    5) Emulation is a hog. If you're going to spend the money on "Big Iron", why not run an OS that doesn't need emulation and gets you every penny you paid for?

    6) "We'll just create another VM for this new process." Managers and accountants see 10% idle time and figure that there is plenty of room for more "stuff" on the server.

    Just my opinions.
    dmhunter@...
    • Oh really?

      "4) IBM used to be really, really big in the IT industry. They aren't quite so big today."

      Um, not so big?

      Intel $37.6 billion
      Microsoft $56.8 billion
      HP $104.3 billion
      IBM $103.6 billion

      What you probably mean is IBM's share of the market isn't what it used to be. I think it's more IBM failing to capitalize on market trends than "why people left" (though changing suppliers is part of the reason for erosion of share.)
      lostark98
      • IBM's global revenue

        is derived from over 80 percent [b]services[/b], the rest is hardware (Mainframe and Unix and Intel Servers) and Software. There's a reason why they sold their PC hardware business to Lenovo. It wasn't making enough money.
        jperlow
        • Does it matter?

          Nevertheless, to the previous post's point, IBM is quite a major force in the IT industry (regardless of the sources of revenue.) If IBM sneezes, the IT industry gets a cold.

          For the record, IBM reported a total revenue of $39.3 billion from GTS and $19.6 billion from GBS which totals 56.8% of the $103.6 billion total revenue. Add in the $2.6 billion from Global Finance and I still can't get to 80% of revenue from services alone. (I don't have a breakdown handy of lab services from SWG and STG, but I'd be surprised if it would bring a services number up that high. I could be wrong!)

          That being said, the margins from services are growing while the margins from hardware and software are decreasing.
          lostark98
          • Revenue

            I think you're just looking at US numbers not worldwide. Per NY Times:

            http://www.nytimes.com/2009/01/21/technology/companies/21blue.html?ref=business

            "I.B.M.?s solid profit performance came largely from higher profit margins in the services and software businesses that now account for more than 80 percent of the company?s earnings.

            In a difficult economic environment, I.B.M. has apparently been successful in convincing corporations and government agencies that its technology would deliver gains in efficiency. ?Cost-saving offerings continue to sell,? Mark Loughridge, I.B.M.?s chief financial officer, said during a conference call with analysts.

            I.B.M. reported net income of $4.4 billion, a 12 percent gain from a year earlier. Its quarterly earnings were $3.28 a share, well above Wall Street?s consensus of $3.03 a share. The company?s profits benefited by 16 cents a share from foreign tax credits and the new research tax credit in the United States."

            So I think its Services+Software = >80 percent

            jperlow
          • Keeping it straight

            Yes, [i]software[/i] plus service is closer to 80% of total revenue (you have to add Global Finance to get over 80%) - but that is not what you said in your previous post. You said: "IBM's global revenue
            is derived from over 80 percent [b]services,[/b] the rest is hardware (Mainframe and Unix and Intel Servers) and Software."

            The numbers I used are global numbers and are from IBM's press release of it's 4Q and YE2008 results. See http://www-03.ibm.com/press/us/en/pressrelease/26510.wss

            Also, the NYT article stated that over 80% of IBM's [i]earnings[/i] came from services and software. That's a [i]profit[/i] statement, not revenue. I'd have to dig a bit into the financial statement to verify that number.
            lostark98
    • VM on the Mainframe, not servers

      You missed the key point, this is not VM on IBM servers.
      This is VM on the mainframe, mainframes don't crash. 5 9s is expected, not the goal.

      IBM servers are just overpriced (IMHO) Intel boxes anyhow. Better than most yes, but bang for the buck ??

      Yes MFs are expensive, but when you can run potentionally hundreds, probably thousands of servers in one, the cost comparison doesn't look bad. There are several ISPs running their Unix/Linux web servers on a MF under VM, thousands of sites.
      Supposedly per MIP the MFs used about 40% of the juice compared to Intel servers. So it is also Green.

      The emulation issue, don't know how this setup handles it. But there was an emulation process out years ago, for what I don't remember, but it basically 'recompiled' the object code to allow it to actually run almost natively in the environment.

      Oh and isn't VM an emulator ??
      knudson
  • RE: You're Gonna Need a Bigger Mouse for that Mainframe

    I think it's about time.
    kbapat