“The mainframe is cool again as shops move to consolidate servers and run enterprise systems on a single platform with robust diversification. Some are even speculating that a rebirth is occurring and are talking about Mainframe 2.0. Currently, IBM z/OS supports many open protocols and other recent innovations including support for Java. Get an overview of the z/OS world for Java developers and learn how to deploy a Java application on the new Big Iron.”
The mainframe is cool again as shops move to consolidate servers and run enterprise systems on a single platform with robust diversification. Some are even speculating that a rebirth is occurring and are talking about Mainframe 2.0.
Dude. The mainframe has never gone away. No rebirth required. In banks, financial institutions and large organisations all over the world that have large amounts of data, that data is not held and processed by Oracle, Java, .Net or anything remotely PC based. It’s all mainframe, has been for thirty years or more, and will continue to be until the end of time. Virtualisation? A hot topic at the moment, but mainframes had that done and dusted decades ago. A mainframe is what everyone leans on and looks to when other parts of the IT infrastructure go t*ts up – which it frequently does.
There is no need for a mainframe version 2.0, or Java support. However, being able to run Linux on a mainframe environment does open up a few possibilities in bringing worlds together.
Java support is more important then Linux support IMHO.
Segedunum has a bit of a one track mind. Everything gets piped through “grep linux”.
I’m a huge Linux fan, myself. But I do make posts which don’t include the term “Linux”.
This isn’t one of them, though.
Java support is more important then Linux support IMHO.
What are you going to run Java and various other things from ‘the other world’ on?
Many times OS is nothing but a base for the applications. You choose whatever make sense, and many times it really does not matter what it is (Linux, BSD, Solaris, even Windows). The important thing is the applications. Therefore, the availability of tools, developer mindset matter. Java has the biggest share on that arena, and it works on all those OS’es. so it makes sense.
In banks, financial institutions and large organisations all over the world that have large amounts of data, that data is not held and processed by Oracle, Java, .Net or anything remotely PC based. It’s all mainframe, has been for thirty years or more, and will continue to be until the end of time. Virtualisation? A hot topic at the moment, but mainframes had that done and dusted decades ago.
I am not certain this information is completely true. Many of Sun Microsystem’s customers are banks, financial institutions, and large organizations all over the world. Sun Microsystems does not sell an IBM mainframe.
From my understanding, virtualization was started with the mainframe to provide multi-user support. This was one of the novel aspects of unix at the time. After that, virtualization was not developed further in the mainframe environment since there was no perceived use for it. IBM would certainly not want to develop operating system virtualization for their mainframe systems and OS virtualization is the current trend now, not multi-user virtualization as an informed reader would suppose from the provided information.
I am currently working on migrating data from a mainframe environment to a unix/oracle/java environment and I often wonder if this is a good business decision. I used to laugh at the need for MS Windows to be restarted on a daily basis but this company’s new unix servers need to be restarted weekly to clear out the memory leaks.
I find this article to be very interesting.
“I am not certain this information is completely true. Many of Sun Microsystem’s customers are banks, financial institutions, and large organizations all over the world. Sun Microsystems does not sell an IBM mainframe.”
You right on this, but wrong on the following:
“From my understanding, virtualization was started with the mainframe to provide multi-user support. This was one of the novel aspects of unix at the time. After that, virtualization was not developed further in the mainframe environment since there was no perceived use for it. IBM would certainly not want to develop operating system virtualization for their mainframe systems and OS virtualization is the current trend now, not multi-user virtualization as an informed reader would suppose from the provided information.”
IBM certainly did develop operating system virtualization for its mainframes. It is implemented in 2 different ways on zseries:
* with lpars
* run VM in an lpar with guests running on top
This works with all of the IBM’s mainframe operating systems (zOS, VM, VSE, TPF, Linux, SAK).
We are on the same boat. The only difference is our client saves $ 5 million operating cost per year by moving from mainframe to Solaris/Oracle/Java.
There is more to consider when running mainframe. There are less and less application developers for legacy code. Cost to maintain legacy application will be extremely high 10 years later. Sooner or later, those legacy code will have to be converted to modern technology.
I just had an interesting talk regarding this with our new CEO, who used to be an IBM and Oracle senior executive.
Our clients are mainly banks, insurance companies and goverment agencies. The trend is already there, it just takes time.
Edited 2006-10-21 03:13
Most high costs in a mainframe environment are software licensing fees and (sometimes) hardware leasing fees (many mainframe customers don’t actually OWN their mainframe hardware).
It’s true that there are fewer and fewer developers, but it doesn’t really take all that much to train a new person in. I’ve been coding professionally on Unisys mainframes for over 18 years, and I would say that it can be a simpler environment to learn than Solaris (for example).
“this company’s new unix servers need to be restarted weekly to clear out the memory leaks.”
It’s a bug, fix it. Or are you saying mainframes can’t have software bugs like a memory leak? (a genuine question)
It depends on the nature of the software that you’re writing on a mainframe.
In the mainframe transaction environment that I write software for (Unisys OS2200 TIP and HVTIP), application programs are pre-allocated a fixed block of virtual memory, and that’s what they have to work with. Period. Those limits are known up front.
Dynamic allocation and paging does take place, of course, but that is done under the covers by the OS and the TIP or HVTIP transaction environment, not at the application level.
If I need more space than I have available for a single program module (a fairly rare occurrence in normal usage), I would create another module (in the case of TIP) and have the first one hand off to the second one, storing transition information in either a fixed memory file or (if performance isn’t critical) in a disk file.
Newer transaction environments use a much larger virtual memory space, and presents limits to the user which I’ve never seen exceeded, but we don’t use that newer environment in production.
When the OS is handling such things, and that OS has been pounded on in production environments and refined over (literally) 40 years, the probability of finding a new bug is rather low. It isn’t like the Windows world where APIs and other aspects of the OS are constantly churning — code that was written in 1966 will still assembler or compile, and even the binary format has remained unchanged since the mid 1970’s.
I am currently working on migrating data from a mainframe environment to a unix/oracle/java environment and I often wonder if this is a good business decision.
I’ve seen a few organisations try this, and it invariably ends up being a big disaster or something that is abandoned and put on the back burner permanently. It’s just too big a job.
It all depends on the nature of the application, as well as its complexity and (perhaps) its dependencies on other mainframe applications.
There are mainframe applications out there which by themselves might be relatively easy to port if you only look at the core application logic and its basic files (the one I work on currently could be one of these), but the complexity of the messages that it sends and receives from other systems can make things a lot more difficult (since those other systems are expecting things to be sent in very specific ways, and are generating their own messages in specific formats).
Also, especially in the airline industry where I work where various layers of proprietary communications infrastructure often exists around the application, the hard part is reproducing the functionality found in the several layers of middleware, not the application itself.
A 2-million-line FORTRAN 77 application can be a challenge by itself to convert to a language like C++ or Java, since that probably involves a complete logical rewrite, involves conplete conversion of the data being stored, etc, but if that system also sends and receives three dozen different data feeds…
I am not certain this information is completely true. Many of Sun Microsystem’s customers are banks, financial institutions, and large organizations all over the world. Sun Microsystems does not sell an IBM mainframe.
Large organizations such as major banks and major airlines tend to use *both* mainframe and UNIX server resources.
When I worked at NWA, we had both IBM and Unisys mainframes running our core systems, but we also had dozens of Sun boxes of various sizes (both clients and servers) running other kinds of applications.
We also had HPUX, AIX, IRIX, and Win2000 servers, and we had Solaris, Mac, and Windows boxes as clients in various areas.
Sun could very easily sell a lot of hardware and software to banks and other financial institutions, but my bet is that most of those are still using a mainframe at some level.
When I interviewed at BB&T a couple of years ago, IBM mainframes were still at their core. I know that the same is true for places like Prudential, Travellers Insurance, etc., because I know IBM mainframe folks who work at those places.
W.r.t. multi-user support — mainframe environments have supported multiple users in various ways since the mid-1960’s. I can’t speak authoritatively about IBM, but UNIVAC/Sperry/Unisys boxes have supported both a basic timesharing environment for programmers (DEMAND) and an online transaction-based application environment (which can support tens of thousands of concurrent users) since the mid 1960’s. That’s how airline reservations systems work, for example.
The idea of “virtualization” as presented by things like IBM’s VM have very little to do with that.
“IBM would certainly not want to develop operating system virtualization for their mainframe systems and OS virtualization is the current trend now, not multi-user virtualization as an informed reader would suppose from the provided information.”
Actually, the mainframe is miles ahead of any other platform in terms of OS virtualization, and has been for decades now. The whole hardware is fully virtualized, from processors, to memory, to I/O channels. Actually, nobody runs an OS directly over the iron, they either run several of them over LPARs (Logical PARtitions) or over z/VM.
To illustrate how advanced virtualization is on the mainframe, one can run Linux over z/VM over another instance of z/VM, inside another z/VM, with almost no overhead (this is impossible to do with VMware, Xen, or any other virtualization software available).
‘To illustrate how advanced virtualization is on the mainframe, one can run Linux over z/VM over another instance of z/VM, inside another z/VM, with almost no overhead (this is impossible to do with VMware, Xen, or any other virtualization software available).’
If that ‘almost no overhead’ is the same as on the pSeries then you might as well go software virtualisation like Solaris Zones/Containers (whatever they’re called this week). The pSeries boxes are crippled by the lack of absolute cpu allocation, ie, the processor cores allocated might be on different MCM’s which means you need huge caches (which are prone to failure as the equipment ages – watch out for the Power6 version of the Celeron with reduced cache…), also the memory overhead per LPAR is just silly. I’d have thought they could managed some kind of zip or rle compression on the memory mapping.
Last of all, remember, that linux partition is z/Linux or whatever its called – plus, as anyone who’s run Oracle in zSeries, the threading is pants. IBM will never fix Linux for zSeries, thats what z/OS and AIX are for.
I have to admit to being just slightly mystified by the idea of Linux on the mainframe (or iSeries) since mainframe cpu time (and memory) is so expensive in dollars, and application availability is second tier. The hardware RAS features are obviously great, but I wonder both if other RAS strategies for Linux servers could not approach it at a tenth of the cost and also if Linux itself is really in the RAS league where such a thing makes a big difference.
On iSeries I find it especially weird considering AIX for iSeries is not expensive and also that POWER, app availability wise, is the only platform for AIX but second tier for Linux.
I sort of wonder if it isn’t in a big part IBM marketing, but different strokes for different folks…
Mainframe = Big Expensive Box that was meant to replace cheap manual data entry clerks. Another great idea from IBM. Now, charging lots of money for a Big Expensive Box built out of cheap commodity parts and running software written for free (linux), and calling the whole thing a mainframe? That’s priceless.
Yes, I went there…
LOL. I don’t necessarily agree with you, but you’re funny!
If I hear any more retarded phrases like “Mainframe 2.0” my head is going to explode.
If I hear any more retarded phrases like “Mainframe 2.0” my head is going to explode.
So, why don’t you think about things like “Magnetic Tape 2.0” or even “Drum Memory 3.1”, or “Punced Cards 3.11”?
egin{sarcasm}
Of course, mainframes are dead. Nobody uses them, because you can buy a cheap PC that you can carry around and which makes funny noises with squeaking buttons and clicking mice. And no one uses magnetic tapes today. And no one needs data from last year.
end{sarcasm}
At this point, I don’t see the need for Java development using z/OS, but if there are useful applications growing, why not? 🙂
…a computer that you can’t throw out of a window”
– Steve Wozniak
Economy cars are happy and friendly and come in pretty colors, but a big ugly freight train is still the best bet when a company needs to move a large amount of material reliaibly and at a low cost per unit.
The movement of data is no exception.
Woz was focusing on end-user computing, not back-end computing, and it’s a very different mindset. 🙂
Is it just me, or does anyone else feel there has been a lot of IBM marketing posing as news on OSNews recently? And nearly all of the “articles” point back to IBM’s DeveloperWorks site.. Many of which are poorly written and/or misleading and none of which are in any way objective.
This particular submission isn’t “news” at all.. Instead it is an extremely high level overview of mainframe technology, which has been around forever and is nothing new, with a splattering of mainframe fanboi-ism…
I think we should have a rule that there be no links to vendor sites, unless it is a page talking about a new product or newsworthy change to an existing product. Otherwise, we will most likely end up with vendors waging war against each other submitting their marketing nonsense as news.
While DeveloperWorks is hosted by IBM and usually deals with IBM applications, the articles are submitted by people that work at many different companies, including IBM. I would have to say that its been one of the best sites I have found to get detailed tutorials and information about core technologies for free.
Its funny, IBM are always touting things as new. Someones already covered the Sperry side of Unisys mainframes, but no-ones mentioned the Burroughs A-series systems, they’ve also been running Java with SMB filesharing, NFS, ODBC and all sorts for years while IBM was still messing about with OS/2! So, in total, they didn’t make the first mainframes, they didn’t make the best ones, and they still don’t have the most open mainframes. It’s funny how the best technology is always overlooked when it comes to buying big blue – they must really do big handouts when selling!
I think this is all part of IBM’s plan to migrate customers to Java apps on MVS/zOS so that when the Power6 hardware comes out next year its easier to start migrating legacy IBM mainframe customers onto better platforms. AIX is beginning to depart so far from true Unix these days I think its a safe bet we’ll start to see more ‘mainframe’ type technology crop up in the pSeries range.