Slashdot Mirror
The Mainframe World Is Alive, Even For Those Under 40
willdavid writes with a link to a report by Jeff Gould at Interop Systems, about the definitely-still-around world of mainframe computing, from which he extracts:
"Last week I had the occasion to visit SHARE, the premier mainframe conference, which was held in San Jose just down the road from where I live.
Based on what I saw, there is one thing I can tell you for sure, and that is that Cobol is not dead. And neither is the mainframe.
When I mentioned to one of my friends that I had been to SHARE, he joked that it must have looked like an AARP convention. But this turned out not to be so. While there were certainly a few 60-somethings strolling around the halls, the under 40 generation was also well represented. What struck me the most was not the advanced age of the people but the relative youth of a lot of the software being discussed."
However, it's not all fountain of youth there, either. (Thanks, BDPrime.)
I spent 20+ years as a mainframe systems programmer. VM/VSE. Since then, I've learned Linux et. al. Man would I love to install Linux in a virtual machine. I'll bet it could fly.
The ol' yellow number 2 pencil is still around as well, as is shoe-making, wine-barrel repair, and of course the oldest tool in the book ... the tool.
Like humans all technologies find their place in the universe. Mainframes have their advantages, just would not want one sitting on my lap.
Can somebody please explain to me what the hell a "mainframe" is/was for and why it might be dead?
According to Wikipedia, Mainframes are a bit like supercomputers but better suited to tasks where there's a LOT of input/output data, but not a lot of calculation involved. Payrolls and such.
As far as I'm aware, those tasks still exist today, probably moreso than in the 1970's and 1980's, so why would the Mainframe be dying out? Have regular desktop/server processors advanced faster than demand for this data calculation and thus are now simply adequate or is this article just a bit of FUD to make 'ol timmy look like he's doing his job?
FYI: I'm most certainly under 40. In fact, I'm barely more than half that age, so excuse my ignorance on the subject; the only times I've really heard the term "mainframe" used is in Films, Games and cheap 80's TV shows. And slashdot.
+1 IDisagreeSoHeMustBeATrollOrAnAstroturferOrAShill
I think you mean the "almost 40 generation". ;-)
I actually just took a job in software development on z/OS (the new hip, backwards-hat wearing mainframe operating system). Aside from the impressive clustering capabilities, we've got a lot of new and exciting stuff. (I personally am a big fan of AT-TLS) It's true that the systems are old and the interfaces archaic and painful to use, but the level of configurability and reliability these things offer is staggering. We have a few customers with 100% uptimes in the 20-year range.
My school (Northern Illinois University) actually is one of the few left offering full mainframe tracks in their computer science department, although COBOL was the most painful programming experience of my life.
I'd bet that my meta-group of 50 or so people has a median age of about 33, and while it is still the old dinosaurs who know the most, the definition of "dinosaur" in my personal, 15 person group is around 50 years old.
Mainframe I'd Like to Fsck!
99.999 Up time, speed, number of transactions, precision, hot swapping, 64 CPUs, lower cost to maintain, longer life time.
Hell, I can get a mainframe and put 30,000 Linux instance on it and use it as a cluster, or rollover servers.
PC Servers still aren't as capable as a mainframe, not even clusters.
The Kruger Dunning explains most post on
I was at the last winter conference in Orlando. I would guess the median age of the attendees was somewhere around 40. There's a LOT of Linux going on in the mainframe world (and COBOL has nothing to do with it). The biggest mistake IBM is currently making IMO is they've gotten into bed with Suse. There was a large group from Suse (Germany) in Orlando last February. Again IMO, Suse is an awful Linux distro. Yast is an abomination to work with on a daily basis. I think Redhat missed the boat there even though their Enterprise Linux distribution has support for System 390 hardware. Anyway, the point is that Linux is alive and well and thriving on big iron.
In addition, one of the primary draws of Orlando is Disney World and the other nearby theme parks. The (oops, almost wrote "Teh" there) February conference was held IN Disney at the Coronado Springs (stay in the Cabanas section if you ever go there, for any reason). SHARE members vote on where to hold their meetings. If a majority of those folks were over 60 I doubt they'd continue returning to Orlando every few years.
If you're not familiar with where and how mainframes are being used today then I suggest that YOU are the one who's out of touch with a significant sector of the computing world. Business' needs don't all revolve around iphones, ajax and youtube. Or payroll and accounting, for that matter.
Raw IO power, in our case. With the number of transactions we process per day (financial services - clearing, trade matching, reconciliation, etc) nothing beats the System-i in terms of raw IO in getting the data in, massaging it and spitting it out...and far easier to manage than a server farm, at least for our use. The same vendor that provides our software also provides a JAVA version, but it's not going to handle the 2 billion+ transactions we do in a quarter.
And this software isn't legacy - it's relatively new and updated on a regular basis to take into account developments in the kinds of products offered.
"Horses for courses" as my British friends like to say./p?
The only reason that we still run a mainframe is because the management in place grow up around the mainframe and their underlings would be put out of work if we got rid of it. There is no reason why we couldn't be moving all of its relatively simple programs from the mainframe to a JAVA or .NET other then the fact that we have to wait for all of the current decision makers to retire or just die. The money we waste on hardware components or in turning on software features that are free in most other parts of the industry as well as the time it takes for old farts to get their head around distributed computing concepts is insane. While they spend days writing a program to do screen scraping to get an answer for management "How many people work here?" before eventually conceding they are unable to get the correct result only to have management come to me for a 2 minute powershell script to get the same information from AD. Yes we store things one way in the mainframe and again in AD or SQL Databases because the mainframe people are scared to try and cross the bridge and work with us. They freak out at anything new and worse they don't how the mainframe works. I read all about the Z/9 in an attempt to relate to those bums, I walked over to their side of the hall and in 15 minutes realized the operators don't care to learn how or why something is, they prefer to think of it as a black box. A big black box that takes up lots of room, lots of power and lots of cash. IBM mainframes exist because people who fear change are unable to get off them, not because there is anything fundamentally advantageous about them. I am planning their destruction, a VM that runs on Intel hardware but responds just like a mainframe, it is software that could be sold for nothing and then all the mainframe apps could be moved to it and IBM would be finished, dead toast. I think it is sad you have to pay and enter a code if you want to see more HD space, you cannot just plug in more SAN space, you have to buy it like you would for the Intel side and then pay IBM to let you see it. It is just a revolting way of doing IT. Mainframes are not innovative people, Mainframes are not sex or cool. Mainframes are anti-hacker, anti-explorer, anti-learning. I cannot stress how much they suck.
Respect the Constitution
I question the need of mainframes today. Now, they are great for running legacy programs (such as payroll, etc) but other than for backwards-compatibly, what advantage does a mainframe have compared to say, a server?
Send me your Resume... if getting payed is something you consider "legacy" then I'd be happy to negotiate some legacy pay terms.
LoL: captch is "weekend"
Quality control, quality control, quality control, quality control
Redundancy, Redundancy, Redundancy, Redundancy, Redundancy
Reliability, lots of it.
LOTS of I/O.
Solid VM technologies that makes VMWare appear like the software equivalent of a toddler still in diapers.
Hardware-accelerated crypto, integrated into the overall system design, and not just an add-on card, at least on fairly modern mainframes.
Some mainframes also run dedicated hardware for CRC on data being churned through.
Designing all that into a cluster leads to something that is just as expensive to operate, and still won't have the same reliability as a mainframe environment.
And no, Google's model does not apply here. Google aren't working with data that must approach 100% reliability to the extent that it's possible for humans and technology to make it.
A lower end mainframe can replace roughly 1000 commodity servers, or so I've been told. It consumes roughly 10kW of power and requires only one operator to keep it functional (well, assuming 8 hours shifts, 3 operators). Those 1000 commodity servers will be consuming ~100W a piece, so the overall power consumption will be 10 times higher than the mainframe, and will probably require at least 3 administrators on the clock at any given time (so going with non-overlapping 8 hour shifts, that's 9 administrators). The cost savings could easily justify the expense of the mainframe, assuming that you are an institution that uses 1000 or more commodity servers.
Palm trees and 8
That's correct. Also, look at the retail business. Merchandise management, loss prevention, warehousing and distribution... And we're not talking arcane software packages.
Here's an example: A retail chain has payroll, merchandising, and warehousing/distribution systems, all on the mainframe. The point of sale interfaces with the mainframe as well. A store starts to run low on an item? The mainframe knows because the POS sends its inventory data constantly. The MMS then tells the warehousing system that that store needs more. A pick list is automatically printed by the warehousing system. The warehouse worker picks the item off the shelf, puts it on a conveyor belt which runs through an RFID portal, linked to the mainframe, that then routes the item to the proper truck in the dock so it gets to the correct store - automatically. The truck delivers the item to the store, and the driver enters that into a wireless device which (you guessed it!) tells the warehousing system and merchandise management system that the item has been received by the store, so the MMS always knows the inventory levels in the store. The associate sells that item, and the MMS sees that from the POS data... it also knows that this particular item pays out a spiff to the associate and sends the information directly to the payroll system, which interfaces with a company who handles payroll (like ADP), and automatically adds the spiff to their next paycheck.
Uh oh - the chain is growing and adding new stores, with increasing volumes of data to process, and the nightly batch processing is taking too long... what to do? Call IBM, license another processor... They activate it immediately for you.
But oh no! A disk is failing... no need to worry, because IBM already knows about it and has dispatched a technician to diagnose and replace the faulty hardware.
New versions of this software are being released all the time, and just about every retailer with more than a few stores uses them. These systems are modern. Don't think a big room full of giant cabinets, reel-to-reel tapes and punch cards. Some of the current IBM iSeries (AS/400) models have a form factor that looks more like a PC than a mainframe.
Show me a Windows or Linux system that can do all that, 24x7, for hundreds or thousands of stores.
Why, no, I haven't meta-moderated lately. Thanks for asking!
Now, they are great for running legacy programs (such as payroll, etc)
Payroll is not a legacy application. You still get paid don't you? :)
My point is, even if payroll were 'rewritten' it would still be a suitable mainframe application.
what advantage does a mainframe have compared to say, a server?
They are bigger.
A mainframe more comparable to a server cluster or server farm than a single server.
They feature processors dedicated to IO tasks. They are the kings of data throughput.
They are also big on reliability. Everything is hot-swappable. Everything is redundant. Failover is automatic and processes are rarely even aware its even happened. They have stuff like io multipathing (multiple redundant buses, controllers, etc) and execution integrity -- multiple processors do the same work, the results are compared, and only if they all agree is the computation accepted, errors are thus averted and defective processors and memory can be detected (and hot-swapped), transparent to the running programs and users.
Because they are basically an entire 'server farm' in a dedicated optimized box they also can require less space and power than a server farm of equivalent capability, which is one of their selling points.
I doubt there are any features of a mainframe that can't be obtained by building a suitable server farm, but at some point in some cases the mainframe is markedly more cost effective.
Basically the mainframe and the software it hosts really make the cash for most enterprises - and as a consequence any sensible management are loathe to replace it with something "newer", even if the systems in question are horrible spaghetti nightmares that no-one really understands, and maintaining them is a process of trial and error. Replacement would simply be "cleaning the inside of a tin can", no obvious shareholder value at all in change for changes sake.
Also technology vendors have finally woken up to the fact that the mainframe isnt a dinosaur on the verge of extinction - for example making CICS transactions web-service enabled has made COBOL code just as capable of participating in a service-oriented architecture as a set of AXIS hosted java classes.
Serious question - I am long out of touch with language design and practice - which currently used and popular language/compiler/optimizers have solid support for BCD and financial arithmetic?
sPh
There's a *lot* of mainframes at IBM-Boulder. They were deploying brand new (at the time) z9's to replace old 360/390 and earlier zSeries. If I recall the conversation with the facility manager for that project, it was a 5 to 1 ratio of old systems to the z9's, most of which would be running Linux VM's for WebSphere deployments of various types.
Exactly. I believe the last reported peak was in the 2200/sec range. And we expect our volumes to double. Fortunately, our System-i (AS/400) team can simply license additional processors 'on the fly' to improve performance. My team even has some of our Unix stuff (I'm the Unix Manager) running in a logical partition on the System-I. And it's freaking fast.
Cobol is still around because it's an extremely verbose language that is easy to understand (I never said "write into"), so managers can understand what their programmers are doing.
z/OS is an upgrade of OS390, but yes - it has something called UNIX System Services, which is POSIX compliant but not as friendly as LINUX.
It's not that z/OS fights changes, exactly. Windows is crufty because it has to run MS-DOS programs from the eighties. z/OS has to run programs from the sixties, and do it with a high degree of reliability.
Disclosure: I am an IBM employee and the author of a mainframe book.
-- Support a free market in the field of government
Mod parent up - he's on the mark. There's a lot of stuff out there for which the source is gone, but it still does the job. Witness the State of California's payroll system we were discussing a bit back.
I was taught to respect my elders. The trouble is, it's getting harder and harder to find some.
The last UNIX system I built was handling over 130,000 transactions PER SECOND.
HP Superdome + Itanium = incredibly fast.
None of the business needs you described sound like something that couldn't be done by just about any UNIX-like OS with appropriate applications (Oracle Financials, JD Edwards OneWorld, SAP). Even the hardware failures you mentioned can be handled with clustering and having cold, warm, and hot spares on hand. And just about any enterprise SAN or NAS has a phone-home feature.
So again, what's so special about mainframes? You guys talk about the bandwidth, what kind of throughput do you get? How many megs per second? Let's put this in raw numbers, I'm honestly curious about why those systems are supposed to be so special.
God invented whiskey so the Irish would not rule the world.
My Dad was a mainframe operator when I was a kid. We're talking early 70's here, but he actually went all the way back to the IBM 1401. I've always had a fondness for the beasts, but no experience with them. Therein lie the problem. They're not at all commodity iron. Linux came into existence because commodity equipment became powerful enough to host such an OS. That cannot be said of z/OS. It simply doesn't run on anything I'm likely to find on sale at Fry's.
How does one gain employable skills with untouchable hardware? (Note: I don't consider Hercules to be a solution. Where's the software?)
Dinosaurs are extinct. IBM mainframes are more like alligators or crocodiles or sharks... mostly unchanged from, and still closely related to, it's dinosaur-era ancestors, but still alive because it's so effective at what it does. Like those animals, mainframes still are the undisputed ruler of their part of the kingdom.
When we're old... hell, when we're dead... we'll likely still have something like Slashdot, with people saying "the mainframe will die any day now".
Life is hard, and the world is cruel
"It is amazing how few people realize how much of our society still uses steam. You forgot geothermal, and some forms of solar plants."
Yup. Back when I was in the Navy, and I got to my first ship, a nuclear carrier, I thought nuclear power was this gee-whiz technology; if you don't know any better, you'd be inclined to think that "nuclear power" is turning those propellers... via protons, electrons, ray-guns, something sci-fi like that. When a nuke machinist mate explained how it really worked, I was kind of shocked. Basically, all we had was a steam engine, where the heat came from a reactor instead of coal. Same process, just a different heating fuel. Since those days two decades ago, I've become fascinated by just how much "advanced technology" that we use is really nothing more than barely improved methods our ancestors used. Jet engines? Nothing more than prop engines with the fans on the inside and some ignited fuel in the exhaust. Ultra-modern rifles like M-16A4's and AUG and the L86? Working from the same priciples as rifles hundreds of years old. Hell, car engines are noting but a hunk of iron with a series of explosions in them.
Technology most of the time is really nothing more than a machine that takes advantage of some principle of nature, and is often very, very simple at it's core.
Life is hard, and the world is cruel
I'm 24 and work in the banking and insurance industry. I spend most of my day looking at green text on a black screen and its not going to change anytime soon. We have two big hulking IBM zseries mainframes running batch jobs all night. We also have a good few servers running unisys, unicentre and techscheduler. I know by far I prefer OPC running on the mainframes, if something breaks its just a bit of bad code or input from inhouse. The hardware itself just never gives any trouble. The users don't care what is running on the background, we have csi's running against cics sessions so all they see is a nice little website with no clue what going on in the background and they shouldn't care anyway. Noones going to go to the expense of replacing all the cobol in the background as long as IBM keeps developing and supporting mainframes. They just work.
Disclaimer: I manage the DASD (disk based storage) for a mainframe (z10) environment.
We used to run a System 390 (think watercooled and took up 1/4 of our datacenter floorspace)
We then migrated to a S390 (aircooled, the first big black box) in about 1995
We then migrated to a z900 in 2002.
We've just completed migrating to a z10.
The applications continue to run on our brand new mainframe unmodified since the early 90s. Sure we've gone from Bus and Tag to ESCON to FICON (1Gb, 2Gb and now 4Gb). But the same applications are taking up significantly less power and a percentage of the mainframe than previously did.
In the early 90s, 486dx was king. How many applications that ran on 486dx DOS systems are currently running in a VMware VM that is running MSDOS 6.22?
Next time you swipe a credit card, and within a few seconds you're approved for your transaction, including determining if you're purchasing something within a common pattern of recent purchases. How many x86 boxes would you need to manage this?
Thank us mainframers.
Btw: I'm 36, run linux at home and on the mainframe.
Most of the reasons have been listed by other people, but I figured I'd put them all in one place and then add a few things:
1) Reliability: 5 9's (99.999%)
2) Backward compatibility, there are people still running applications written 40 years ago
3) Security: Physical (hard to move a refrigerator), Network (no external network when applications working internally), RACF, Highest level of security rating of ANY server, ever.
4) Architecture: Redundant everything: Spare processors, spare power, spare, everything. Predictive failure/automatic fail over for individual components. Memory Bus greater than anything out there. Pipes to Storage extreme. Cryptographic processors to do SSL, etc.
5) Scale up: 64 processors (4.4GHz), 1.5 TB of Memory, etc.
6) Scale out: GDPS (Geographically Disperse Parallel Sysplex) up to 32 boxes?
7) Hipervisor: Its a network in a box. Applications talking to each other use IP, not TCP/IP, so you aren't sending 35% data, 65% header when applications talk. Network is at the speed of memory. zVM has been developed for over 20 years.
8) Power Efficiency: Compared to a server cluster + cooling + redundant power, etc.
9) Network Simplicity: No need for a rats nest for your rack, cable simplicity in some cases from over 1000 cables down to 12. From 14 switches (which are very expensive) to 4.
10) Management simplicity: Less staff needed to keep it up and running. Instead they are focused on adding business value
11) Running Legacy (aka Business Critical) applications, your web presence, your portal, and a myriad of other disparate applications in one place.
12) Create new servers in minutes without needing hardware "on standby."
13) Compartmentalization in a single box
14) Shared everything while still fully separate
15) Workload manager: able to on the fly change how much resources are allocated to images AND (this is the cool thing, cause other VMs do that) give it goal times for operations. As in: Complete this task in 1/100th of a second, and it will allocate, on the fly, for that to happen, and it will guarantee it.
I think that's enough...I'm sure I'm missing some stuff, but that should be a good compilation.
One thing to note: I'm under 30, and didn't know that mainframes existed 5 years ago.
Mainframes are NOT the answer to all questions. Intel is NOT the answer to all questions. Itanium, Solaris, Power, etc...none are the answer to all questions.
Buy the right tool for the right purpose.
In the 90's, the AS/400 went from a 32-bit CISC processor to a 64-bit RISC processor.
If this were Intel, we would have had to go through the code looking for bustage (like we did when Windows' wParam went from 16 to 32 bits). This being an AS/400, we didn't even have to recompile! Just copy everything to the new system; on first execution, programs are converted to 64-bit opcodes.
ALL AS/400 (iSeries, System i) programs have been 64-bit since the mid-90s; no programmer intervention required.
"The empty vessel makes the greatest sound." -- William Shakespeare; Henry V, 4. 4
And scaling up a single mainframe gets exponentially more expensive; the price buying more commodity servers stays constant, meaning a linear cost of scaling.
Why do IT people (I'm assuming) make such lousy economists?
Think about it for just half a second. If you want to double the transactional capacity ("throughput") of your mainframe, you turn on more processor(s) inside the same box. (It's almost always inside the same box. A single mainframe can have massive capacity.) And you only do so when the mainframe is 100% busy for sustained periods, which it gracefully handles by the way without choking. You have zero reconfiguration to do to hardware, software, or applications: it's "on tap." (In fact, the machine itself can provision itself nowadays.) This is very cheap: doubling costs way, way less than the initial allocation. Also, up to 32 machines can share memory and operate as one, so if you are unusual and hit a single machine limit, no problem. It's only past 32 that you resort to partioning, traditional clustering, etc. -- and you've still got a lot more weapons at your disposal.
If you want to double your actual transactional capacity with highly distributed servers, you...well, it may be impossible. You better hope you have highly segmented and partitioned work for those servers to do, and that it is extremely well balanced so that it fits into little server buckets. The burden is mostly or entirely on you, the application developer, to figure that out -- and that's horribly expensive. (Because the work never is balanced, you have to install a bunch of servers that don't run continuously at near 100% busy, so you get less real-world performance out of each processor anyway.) But at the very least you have to more than double the number of boxes. And you have a lot of knobs to twist and turn to get your work settled into its new, large number of machines, so you better hope you don't need that extra capacity NOW. (Can't process those credit card transactions during a spike in demand? Sorry about that -- I guess your credit card company will just have skip collecting those millions of dollars.) And you get to hire and pay some more people to install, configure, and maintain those extra boxes. You also get to find more space in your data center (if you have it), consume more than double the power (if you have it), and run the air conditioning more than twice as hard (if you can). You pay Oracle, Microsoft, et. al. at least double, too. (That's not how mainframe softwre works: there are strong price curves, not lines.)
Fortunately there are some IT-economists in the world who understand this stuff.
Actually, IBM has been putting a lot of POWER inspiration into its z10 processors, although they are not at all the same. That means a z10 core is clocked at 4.4 GHz. It also has hardware decimal floating point, something nothing else has (except POWER6), which speeds floating point calculations by about 3 orders of magnitude versus software.
These processors are state-of-the-art, even in raw computational performance terms.
A company I used to work for after being bought out implemented SAP because that was what the new owners used. What did 6 years and $300 million get them? Less features, and stock inquires went from ~2-3 seconds to about 10 - 30 seconds. Or several minutes if there was a problem. At the time, the mainframe was running about $3 million/year including support staff. Thanks to SAP, I now snicker every time I see car commercial touting "German engineering"
IBM once made a desktop mainframe, the PC/370. You could run VM on it. But that was in 1985. Since then, they've avoided offering low-end mainframe compatible machines. There's no reason IBM couldn't offer a 1U server that runs zOS for $2000 or so, but they don't. Remember, most of the software was designed to run on machines well under 100 MIPS.
As other people have pointed out, IBM-type mainframes do virtualization right. Virtualization on x86 is a hack and an afterthought, even with the newer hardware support. x86 virtualization with VT hardware creates a virtual machine that doesn't look like a bare machine with VT hardware; the virtual machine has no "ring -1". VMware actually patches code on the fly to work on older x86 hardware, which makes VMware very complex and vulnerable to bugs. The mainframe people don't have do that. On IBM-compatible mainframes, the virtual machine can look just like a real machine; you can run VM under VM under VM, and it works fine. About ten deep, it's too slow to be useful, but it works. This is good for stability.
For historical reasons, PCs have a primitive I/O architecture. The "bus" concept came from the days when the peripherals and the memory were really on the same bus. That hasn't been true for decades now, but the architecture is still set up as if it was, with peripherals seeing physical, rather than logical addresses. In mainframes, there's an I/O MMU, memory protection between the peripherals and memory, and there's a channel architecture which standardizes how peripherals talk to the computer. PCs are still stuck in the "each peripheral has its own device register layout" era, which is why we have so much trouble with drivers. The "device register" and "bus" concepts are so deeply embedded in PC thinking that FireWire, which is really a local area network, was designed to emulate a bus with device registers.
a dailywtf.com story in the making.
Comment removed based on user account deletion
A z10 has 336 I/O controllers. Each one is a ppc processor. You do the math.
The last time I saw a zSeries tape drive, the tape controller was a pSeries.
There's no failure quite as dissatisfying as a complete and total solution to the wrong problem.
Has anyone actually done this? For me, Google returns exactly one hit -- your comment. Nice googlewhack, but not exactly a valid citation.
(ooohh, you mean without the quotes!)
"Good news, everyone!"
In the 90's, the AS/400 went from a 32-bit CISC processor to a 64-bit RISC processor.
If this were Intel, we would have had to go through the code looking for bustage (like we did when Windows' wParam went from 16 to 32 bits). This being an AS/400, we didn't even have to recompile! Just copy everything to the new system; on first execution, programs are converted to 64-bit opcodes.
ALL AS/400 (iSeries, System i) programs have been 64-bit since the mid-90s; no programmer intervention required.
...because the compilers available to application developers generated code in a very high-level pseudo instruction set that the low-level system code translated to native code on first execution. The high-level code was (usually) kept around, so that on the first execution on a RISC system it could be translated to the (extended) PowerPC code the RISC processors run.
The System/360-to-z/Architecture machines didn't do that. 32-bit data/24-bit address programs stay 32-bit data/24-bit address programs even when run on 32-bit data/31-bit address or 64-bit data/64-bit address machines, and 32-bit data/31-bit address programs stay 32-bit data/31-bit address programs even when run on 64-bit data/64-bit address machines.
However, the processors can, at least, still run those programs (just as, for example, x86-64 processors can run IA-32 programs, or SPARC v9 processors can run SPARC v7 or SPARC v8 programs, or..., OS permitting).
I still miss having my Unix command line thought!
And I'm 34.
--
Luck is just skill you didn't know you had.
IBM is in the process of unifying its processor designs to keep costs down. The CPUs you find in their high-end POWER boxes and the mainframe systems are almost the same, but with different instruction decoders. The POWER 6 has a relatively modern instruction set (a superset of the PowerPC instruction set from the '90s), while the z10 is binary-compatible with the OS/360 machines from the early '60s. The POWER6 is currently the fastest CPU on the market for raw number crunching and a z-series mainframe can have up to 64 chips with the same execution units running at around 4-5GHz.
As the other poster said, both the Z10 and the POWER6 come with a hardware decimal floating point unit. In a lot of financial applications, it is important to have a fixed number of decimal digits of accuracy. If you try to represent 0.1 with a binary floating point value, you lose some precision (1/10 can't be represented as a finite sum of 1/(2^n)). To avoid this, it is common to use binary-coded decimal (BCD) values, where each byte stores a value from 1-100 - basically using 4 bits for each decimal digit. If you program in COBOL you can use these directly. In Java and most other languages you can use them via a library. You can perform BCD operations easily in software using a combination of integer operations, but just like software binary floating point this takes a few hundred instructions for each primitive operation. If you're doing the kind of financial operations where BCD is important then a single Z10 or POWER6 will outperform a moderate sized cluster of Xeons.
I am TheRaven on Soylent News
Because that gives you fixed point arithmetic. BCDs allow floating point with a fixed number of decimal significant figures. The POWER6 and Z10 chips have a hardware decimal floating point unit for precisely this reason. Trivial case, consider 0.1. You can't represent this in any finite number of binary digits (just as you can't represent 1/3 in decimal). If you need a fixed number of decimal places of precision, you can use fixed-point arithmetic. If you need a fixed number of decimal significant figures, you can't use binary at all because 0.1 is expected to have the same precision as 1.0 but in reality 1.0 will be stored as 1x2^0 while 0.1 will be stored as 2^(-4) + 2^(-5) + 2^(-8) and so on until you run out of binary digits.
I am TheRaven on Soylent News