IBM confirms new mainframe to arrive 'late in first half of 2022' IBM has confirmed that a new model of its Z Series mainframes will arrive “late in the first half” of 2022 and emphasised the equipment's debut as a source of improved revenue for the company’s infrastructure business. CFO James Kavanaugh put the release on the roadmap during Big Blue’s Q4 2021 earnings call on Monday. The CFO …
I never used a 'mainframe' either,... but a decent sized VAXCluster, and IBM AS/400s. Only place I've seen an IBM Mainframe in the flesh, was at a bank. (they had two, as they had mirrored datacentres). so I think there aren't many niches for such. Never saw one when I was IBM, in IBM datacentres, was mostly P series iirc.
I think there aren't many niches for such
Selection bias.
The bulk of my work is in IBM mainframe environment emulation, so I see customers with z systems all the time. If you don't work with z or its predecessors, you're less likely to run into them. And they're expensive beasts to lease and maintain, so they don't proliferate the way Windows and Linux systems do. But there are plenty out there.
Just look at 3270 emulators. That's software for the particular proprietary green-screen terminal family that z uses.1 Considering that a great many mainframe apps are front-ended with Java or web UIs, or have no UI at all, that's got to be even more niche than the machine itself.
And yet you have free 3270 emulators that are still maintained and widely available, such as x3270/c3270. And you have a wide range of commercial emulators from various vendors: IBM's PComm; our HA Cloud, Reflection, and Rumba lines; bundled ones as in Microsoft's HIS; and various smaller offerings such as BlueZone, Rocket, and Turbosoft. All still viable products. And you have people writing new Internet Drafts proposing enhancements to the TN3270E protocol, and so on.
IBM wouldn't be creating new models if the market weren't there.
Of course, they're heavily virtualized (VM/CMS was nearly 30 years old when VMWare was founded), so it's rare to see a physical system, which might be running hundreds or thousands of LPARs across an organization.
1System i and its predecessors such as the AS/400 use 5250, of course.
Before I retired 22 years ago I worked in businesses that had 2+ mainframes. In fact, the only time I saw just one mainframe was back in the '60s. My last shop had 6 in the main computer room that supported 16 versions of Z/os and had 4 other locations throughout the world. I have interfaced with other professionals that made us look small. We also farmed out our DASD to different locations of the world.
A friend had to use a certain piece of software for his master's thesis research. The first page of a user guide booklet stated that the recommended hardware was a mid-range Cray. The (slightly overwhelmed) student took the booklet, went to the school's IT and asked shyly about resource availability. To his astonishment, they booked him for some ten minutes every other day on a similarly powerful thing somewhere across the country.
When he turned in his results for a routine check with his advisor, the man looked astonished - how did a student get so many results in such a short time? He was expected to run the program on this core duo laptop.
"Work consists of whatever a body is obliged to do. Play consists of whatever a body is not obliged to do."
Anyone who wants to get a taste of IBM 3xx life can try out the free Hercules emulator and the old IBM OSes and other software which is legally available for download and use on it. It's not the same as modern z, but quite similar in many ways, and the look and feel hasn't changed (at least if you still do things the traditional way).
Of course it might not be easy to find your way around. I believe there are various walkthroughs for getting things up and running with Hercules, though, and there are books for aspects of using z-and-predecessor OSes, such as Fake Your Way through MVS.
They're somewhat interesting; MVS being more of an academic interest (at least IMHO: I can't say I really enjoyed using it) whereas VM was reasonably okay as an interactive system: not outstanding, but nice enough. If I had to choose I'd sooner go with a Vax or PDP-10 tho'. Edit: obvs. talking about an alternative reality where DEC is still a thing and the KC10 project succeeded.
It is an odd phrase "power more MIPS (Millions of Instructions Per Second) than in any previous program in the company’s history". Since they are only looking at old IBM hardware, what other companies have products with higher MIPS than this IBM offering ? There much be quite a few, just from the way that they made a statement that is always true. Or a statement that does not provide any useful information at all.
It's a marketing press release. Of course it contains no useful information.
But to be less cynical (as hard as that is), I assume what they're getting at is that they've sold more Z15 computing power than other individual models previously, which is both a testament to the power of their newer Z-series CPU, but also of the fact they've managed to keep sales up.
Each new generation is a step up from the previous one. So you get to run more workload per unit cost of infrastructure, lowering your cost per transaction. It's a marketing piece to existing customers used to thinking in MIPS and in large percentage improvements. I don't have any exact numbers, but gut memory is that Z series was 40% faster than previous generation. The promise in droid-speak is that the new series will be similarly better.
I don't think there's a similar comparison to other hardware types. The workloads tend to be many small transactions of limited parallelism that require ordered processing and persistence, at least that's my narrow experience! :).
Most hardware refreshes are not to get extra processing power, or to gain some mythical lower cost per transaction (who runs their infrastructure at 100% 24x7 anyway?) but because maintentance costs for ageing hardware become significant. I'd be happy to sweat my assets, if support didn't become prohibitive.
I've been with an outfit that upgraded to the most powerful mainframe each and every time one became available from IBM (and once very long ago from Amdahl as it was more powerful ). We just needed the raw processing power despite having many thousands of distributed servers around it.
We worked 24X7X365 as well as all our mainframes. If we were down for anything more than 30 minutes we were fined by the Fed at 10,000.00 a minute in my 8 years at the company we were fined once and heads rolled. One time a program malfunctioned (Not written in-house) and we were hit big time. Needless to say, the OEM software program was eliminated in a week.
I was similarly amused at the wording on some cleaning product which said in huge letters that it's 10X better than something else, and in tiny letters that the something else in question specifically does something different. Along the lines of, "DOG! 10X BETTER* than cat!!
* at being dog."
A client I'd consulted for on an SAP migration that involved complex EMC array setup would need a similarly complex setup for a migration to a newer mainframe.
Because I was already in discussion about an opportunity at Goldman Sachs and loved the idea of weekly commutes to NYC in a paid for apartment (where instead of going home I could have them pay to bring my girlfriend out for the weekend) and because I figured "for how much longer would learning about IBM mainframes be a worthwhile skill" I turned them down.
I instead focused on the Goldman Sachs thing that once everything was settled was scheduled to start a couple months later on October 15. October 15, 2001. Ooops! Coulda woulda shoulda!
It is irrelevant to compare MIPS between different platforms. MIPS are the low level instruction a CPU and operating system execute. Different CPU types, IBM, Intel, do not have the same instruction set. For example a simple add instruction could be simpler and more efficient on one type of CPU than another. MIPS are only relevant when compared on the same platform. In this case Z series. Also when the operation system changes, the MIPS rating can also change. Some operating system changes can either improve or waste CPU cycles. A lot of this really depends on how the vendor actually does the testing and rating. Some smoke are mirrors depending on how honest they want to be. I worked at Tandem computers and the horrid companies that bought them years ago, Compaq and HP. Occasionally did performance and tuning for customers. Even benchmarks.
Not a problem, since IBM mainframe "MIPS" don't mean MIPS anyway, and haven't for a long time. They're a metric IBM derived years ago when z started getting architecturally more complex.
So they're only useful for comparing different z models, or figuring a rough estimate of workload size.
25 years ago when I last worked at a large Fortune 500 company nearly all the main business processes were running on a few S/390 mainframes. Online and batch.
It's a testament to the operational efficiency of Z/OS, MVS etc. Because it's all consolidated in one place it might well be cheaper than thousands of disparate servers and teams designing and managing infrastructure. I now make my living do the latter, so better not bite the hand the feeds me!
IBM mainframes and software are as close to bulletproof as you can get.
They are also a major upfront cost in a world of beancounters that are turning to cloud solutions. When this IT age is analyzed, the phrase "penny wise and pound foolish" is sure to be use (if anybody remembers what a pound is).
If I were a CEO (fat chance) and I could utilize the MIPS I'd run my company on IBM mainframes.
This "power more MIPS (Millions of Instructions Per Second) than in any previous program in the company’s history" sounds stilted because it's mainframe terminology. Similar to how DEC would rate their systems in VAX MIPS, IBM specified some level of performance as "1 mainframe MIPS" decades ago, and you can pay for a system with a certain number of MIPS. It has spare CPUs both as hot spares (even in-progress work, the mainframe CPUs run two side-by-side pipelines, any mismatch is detected, the jobs running on it moved live to a spare CPU) and as additional capacity (If you find you need more MIPS, you call IBM and they turn on more CPUs).
I think it's a nice way of saying they're selling fewer mainframes (I doubt it's a growing market) but the remaining mainframe customers are putting more load on them and so ordering more MIPS than before. Would not be surprised to find quite a bit of this is mainframe Linux (... if you have a mainframe and some Linux servers accessing the mainframe and obviously providing result via TCP/IP, you can move those Linux workloads onto the mainframe itself. ) A bit bizzare, but IBM charges a significantly lower price to turn on a CPU for Linux-only use versus a "full CPU"... I think they disable a few instructions that only the mainframe software uses but Linux (and GCC & Clang compilers) don't; so if you try to cheat and run DB2 or whatever on it it won't start.
Not to dissemble, but mainframe MIPS are odd; they measure the number of instructions run by the CPU, but mainframes have IO processors that (for the mainframe databases and record stores at least) do a major amount of the work that would normally be done by the CPU; a single IO instruction could pull a database record into memory and provide a pointer to a particular field in the record.
Edit: I can't put my finger on it but it seemed ironic that IBM's hoping to boost sales by putting out systems within 3 years, when these systems will last for decades with proper maintenance (and part of the hefty fees to IBM are for proper maintenance.)
The issue is replacement parts. No vendor wants to keep, build or repair (if possible) old parts. And by old I mean perhaps 5 years. Think of hard disk drives. The capacity increases so quickly that a 5 year old disk is obsolete, unremarkable and no one will build 5 year old tech.
I loved mainframe more than Linux & Windows!
We converted patented utility to Java on IBM mainframe four years ago, but never got to specifically play with “free” built-in Parallel CPUs for array-processing and encryption and native JSON support. Had brought in Netezza MPP database before IBM bought and rolled in for M/F DB2 to automagically exploit “transparently behind scenes under the covers” decided by cost-based-optimizer unbeknownst to average programmers. IBM Mainframe Z/DB2 is still my having fav XML database support. Last z/OS & M/F had fastest CPUs on planet & best “hyper visor” for workload/Thruput management w/best logging/journaling for cross-cutting concerns we had to build-into midrange deliveries. Last-century Software AG 4/GL Natural enables us to “phase-in” fixes to CICS/DB2 applications into running production like much touted CI/CD today. Mainframe “compiler” externalized “everything Compiler knew” (SYSADATA data and heuristics from procedural & object-oriented COBOL/ASSembler) which I loaded into Teradata & DB2 tables for analysis “across” programs which I always wanted to use to auto-generate other language manifestations. Seems if ANYBODY can implement “online mobile voting” for 2024 elections via crowd-open-cloud computing agile delivery paradigm it would be this solution-set with end-to-end encryption and abilities to aggregate on encrypted & compressed data.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
