> And, also, at the end, the article mentions that Fujitsu is finishing with mainframes in 10 years.
I think people are reading more into this announcement than it actually means.
Fujitsu has three mainframe lines: GS21, BS2000, and VME. The announcement it links to is only about GS21 hardware, not BS2000 or VME.
GS21 started out as clones of IBM mainframes, including both Fujitsu's Japanese clone line, and also Amdahl in the US. In the 1980s thru the first half of the 1990s, Fujitsu's Japanese clones were popular in some countries outside Japan (e.g. Australia), and Amdahl's clones enjoyed a lot of popularity in the US. However, while IBM's own mainframes went from 31-bit to 64-bit in 2000, Fujitsu's GS21 line got stuck on 31-bit. AFAIK, nobody uses it anymore except in Japan, non-Japanese customers either migrated off the mainframe or switched to IBM. In Japan, it clings on in government and some large corporations, but is deeply legacy.
BS2000 has a very different heritage. RCA used to sell mainframes in the US. One of their (several incompatible) mainframe lines was partially compatible with IBM S/360, having the same assembly language in user space. However, kernel mode instructions and the software were incompatible. Siemens in Germany bought the rights to it. It ended up surviving in Europe even after it died in the US. Eventually Siemens sold it to Fujitsu.
VME was originally the British mainframe manufacturer ICL, which Fujitsu ended up buying. The British government is (I believe) the only people still using it, and they've been trying to get rid of it.
Both VME and BS2000 haven't had their own physical hardware for years now – they are just software emulators running under Linux or Windows. As such, they are easy to "migrate to the Cloud". Fujitsu is likely to keep on supporting them (to some degree) as long as there remains at least one customer willing to pay enough $$$ to sustain it. Given that, any timeframes for phasing either out should be taken with a large grain of salt, since their future really depends on how long it takes the remaining customers to get off them.
Coming to this GS21 announcement – I understand Fujitsu wants to stop manufacturing and selling physical mainframes in Japan. They can just switch to software emulation like they do for BS2000 and VME. Like both of those, it can run fine in the cloud, and I expect they'll keep it running as long as there is customer demand. Promised future support timelines are very notional, if there is a big enough customer paying for an extension, the extension will happen. Plausibly, it could still be supported in 20 years time – maybe it is used by some essential Japanese government system, and the Japanese government's repeated attempts to migrate to a more modern platform end in failure (not an uncommon story for government IT projects)
I haven't kept touch with the mainframe industry, but yes, you captured it succinctly, last I knew, a lot of IBM mainframe stuff was running on emulators. Albeit on highly available hardware.
> last I knew, a lot of IBM mainframe stuff was running on emulators. Albeit on highly available hardware.
That's rather different from what I've heard: IBM refuses to allow any remotely recent versions of their mainframe operating systems to run in production on anything except their own physical hardware. A lot of hobbyists run ancient OS versions under Hercules – like MVS 3.8J, which has bugfixes up to circa 1985, but feature-wise is stuck in the early-to-mid 1970s – which is legally allowed since IBM intentionally released those super-old versions into the public domain. But, I doubt anybody is still using OS versions that old in production – maybe there's a handful of sites still running 1990s versions, but all of those IBM won't let you legally run on Hercules. Basically the only emulator they allow is their own super-expensive one (zPDT/etc), and only for development use. Hercules can run newer versions fine, but not legally. [0]
So, while almost every other mainframe vendor has moved (or is moving) to a purely software-emulated solution, IBM insists on physical hardware – because they are trying to keep their high-margin mainframe hardware business alive for as long as possible. Eventually, as the install base continues to gradually decline, they'll likely terminate the physical hardware line, and pivot to software emulation as other mainframe vendors have. However, (just like all the other remaining mainframe vendors do), they'll only license the OS to run on their own proprietary emulator, not open source ones like Hercules.
One option some sites have gone with is "application-level" emulation aka mainframe rehosting software – don't emulate the actual mainframe hardware to run the real OS, instead recompile the app using a COBOL (or PL/I or whatever) compiler for Unix/Linux/Windows, combined with an emulator which just emulates the mainframe OS calls (and calls to middleware products such as CICS). This approach is taken by products such as NTT/Dell's UniKix and Oracle's Tuxedo ART. The downside is you need to recompile your app on the new platform, and the emulator only supports a subset of the OS/middleware calls. If you are lucky, and your app is pretty vanilla, it may all just work. If your app relies on obscure or newer OS/middleware features, you may have to rewrite significant portions of it. Also, it doesn't address large parts of the skills gap – you still need to know COBOL and mainframe APIs, but now you also need to understand what subset of those mainframe APIs is actually implemented by the rehosting package – although at least OS-level admin is now mainstream and the exotic stuff is purely application/middleware level
[0] With some super-rare exceptions: e.g. IBM used to allow people to run their operating systems on FSI's FLEX-ES, mostly just for development, but including some low-end production use. From what I've heard, they stopped allowing this for new customers 15+ years ago, but sites who licensed it for low-end prod use back then are allowed to keep on running it, and there are still a small number who do. I've also heard that those customers are stuck on really old mainframe OS versions – although late 1990s/early 2000s rather than the 1970s/1980s of the public domain MVS 3.8J – because IBM refuses to license newer versions to run on FLEX-ES
Yeah, I'm also curious how many total jobs this is.
And also what the pay is. These jobs are overall more important than the tens of thousands of random software engineers at FAANGs.
To solve the talent deficit, they could say "We pay like FAANG. And if you do serious work, we expect stable jobs and steady career progression for at least 20 more years." Fund it by investing some of the wealth where it needs to be invested.
(The hard part I see is holding companies to career progression, so they don't just bring in people at Google L4/L5 equivalent TC, but plan for those people to become less mobile to other employers due to mainframe skillsets, so easy to retain without normal tech growth in compensation over time.)
As a former P-series/AIX specialist, I can say that the pay is not as good as a decent linux job, and the career growth potential is very low. Your career options are basically to maintain legacy systems until they can figure out how to migrate it all to linux.
And if you get laid off (happens a lot at IBM lately), your career prospects are to either wait for someone to retire/die or learn linux.
No matter how important they say maintaining the legacy systems is, the pay is low enough to easily convince people to take any other IT career path.
>These jobs are overall more important than the tens of thousands of random software engineers at FAANGs.
the jobs are cost centers at those places while at FAANG it is main R&D, areas of investment, and thus very different treatment by the management, incl. compensation.
One can also wonder how long before LLM would replace the [most of the job of] sysadm of mainframe. I'm pretty sure that the management at those companies would jump at the mere perception of the opportunity to shave a bit of cost like they did with outsourcing.
There are active efforts at IBM to market generative AI to rewrite mainframe code (COBOL->Java). Orgs are desperate for cheap mainframe folks to bridge the gap until they get off of mainframes while most of this skilled workforce is retiring, but the comp and work arrangement quality is likely subpar for a job orgs are actively attempting to eliminate. If you’re close to retirement and have the skills, it’s a fine way to ride into the sunset, but if you’re younger, run away. It’s a trap.
All the Pain an unclear future of the mainframe is IBM's fault. Its insanely closed ecosystem, from hardwar do software. No one other than IBM has access to it. Seems to me that going further not even IBM team will know how to manage.
Once market builds an interesting compelling offer to mainframe, enterprises will leave IBM as fast as possible.
To deploy mainframes, you've most likely spent millions, sometime might even be hundreds of millions. In other words, money is not the problem. So why would someone ever risk getting into huge trouble by reverse engineering or cracking?
I’d gladly work on this stuff if it were actually lucrative.
There’s a secondary problem: they literally just don’t make things the way they used to. Nearly 100% of all SPAs i visit are busted in some manner, or break depending on the day of the week. Main frames and the culture that surrounds them is one of not breaking, ever, and maintaining backwards compatibility into eternity. There are js modules that measures their lts support in months. Most actual engineering has been lost out of modern software development: instead we just throw whatever was last month away for the latest redesign.
In around 2015 or so, prior to social media, etc, there used to be a mainframe forum or two (perhaps they still exist) where a whole bunch of newbies from India used to hang out, to learn and grow their mainframe skills. It is the same time when there were stories floating around of mainframe veterans being let go. People have short term memory issues.
If IBM weren't hostile to the Hercules project and allowed local licensing to run z/OS, CICS, IMS and DB2 on it, perhaps more hobbyists would want to careerpath themselves on to the s390 architecture.
I do love the s390 arch and the massive IO hardware over there, but IBM has paywalled down entry so hard that there is no audience.
They even went to the trouble of making Go binaries transportable for direct execution under z/OS. But if you want new people to write code on the platform you need to make access to the platform a thing.
Former P-series/AIX SME here. I agree 100%. IBM's training programs are a joke.
P-series and mainframe machines have a lot of cool tech, and they're very resilient. They can even lose a CPU or some RAM and keep running. x86 systems would more likely freeze/crash immediately.
But the only reason I know P-series/AIX at all is because one small branch of IBM hired me for my linux skills back in 2011, and I learned on the job. But I quit after 5 years, because the pay wasn't sustainable. The machines are too expensive to play around with otherwise. If you learn by doing (which seems vital to be a good sysadmin or programmer), even a license to use AIX is out of the hobbyist's price range. Training courses are limited lab environments. You won't get nearly as much out of that as you would from a 12-month AWS subscription, or a $5/month VPS, or an x86 virtual machine, or a raspberry pi. etc, etc.
And IBM ended their developer machine licensing. So now employers can't even afford to maintain extra P-series machines for devs/sysadmins to play around with and learn.
But don't worry, IBM will keep shooting their feet off until they no longer exist. There will likely be a panic, similar to Y2K, where everyone's feverishly re-writing and porting and emulating and migrating things off of IBM iron and onto x86 machines.
That is a very good point. Why buy big iron "pets" when you can buy x86 "cattle"? It works for a lot of stateless apps and services. A node dies, and k8s quickly moves the workloads to new nodes.
It does take some work to accomplish this on more complex apps, though. Things like SQL databases and rabbitmq are very often single points of failure in practice. At smaller companies, it's often easier to stick them on more resilient hardware than to architect an active/active or active/failover system. I agree that this isn't the best way to do it, and IMHO any important service should have HA of some sort.
That said, I use x86 based machines for all my personal projects, and I wouldn't buy IBM systems if I owned a tech startup just because they're like 10x the price of x86.
IBM: Footgunning before the XT and before the Republican conformity dress code when it was cool to enable genocidal regimes for money. IBM is the quintessential big, dumb company that innovates in spurts and then is a perpetual loser because of all of its corporate bs.
And now that they bought RedHat, they're eagerly dropping support for all linux distros that are not RedHat, and they're attempting to lock out anyone who wants a personally affordable RedHat-like system (CentOS...). If they succeed, it may even be hard to find decent RedHat sysadmins in the future.
It's hard to keep up a trained workforce when they lock the doors so tight.
Yep. In the greedy drive to monetize the Meta's and Motorola's running Cent to force them onto RHEL, they're pulling a grenade pin and daring people to stay in the boat. Instead, they're alienating most of their future business of potential users and decision makers by attempting to emulate a worse than Oracle while disrespecting their users.
The value in something like Cent/RHEL is LTS stability. Really, the only valuable part of Cent/RHEL is the stable kernel because the userland rapidly becomes old and useless. Something like Ubuntu with a RHEL quality kernel with a 10 yr support lifecycle but without a separate foss/commercial version would be superior to either RHEL or Ubuntu. Canonical went full shark-jumping mickeymouse with their "Pro" subscription and non patching model unless you pay $.
The mainframe is not a special thing anymore, hasn't been since the late 90s. It's just a server box.
I work at a shop with a z/14. I would love it if we finished the last COBOL retirements and go back to the mainframe but this time to run container farms, fresh Go code, and use thr power to run way deeper matrices of tests that take days to run locally and cannot afford to run on AWS.
IBM could sell the future of the on premises z/xxx boxes as "datacenter in one rack"
Running x86 in z/VM has been a discussion for 25 years. Just fucking do it. Let people run whatever they want.
Just as people are excited about ARM for low-watt computing, make s390x just as exciting for people who want insane vertical resources but using the same dev tools that are used now for easy x86/ARM crossover.
But IBM culture has always been about overcharging a small and rich audience and now they are sitting around hocking their services cohosted thru AWS and everyone who has COBOL and 360ASM running are doing retirements with no plans to use the boxes after its all unloaded.
> Just as people are excited about ARM for low-watt computing, make s390x just as exciting for people who want insane vertical resources but using the same dev tools that are used now for easy x86/ARM crossover.
The funny thing is that AIUI the technology is already basically there. They actually did throw the resources into getting a lot of Open Source software to be compatible with s390x, they've got Linux LPARs and LinuxONE. My understanding is that they just... don't make any effort to sell, outside a tiny fraction of Enterprise™.
Performance per dollar and per watt of IBM machines are wholly out of step with the market. As such, there's no motivation to re-use those machines after migration off z/OS or other IBM solutions anyway...
There is a store chain nearby hiring for an IBM mainframe dev. It's unlikely anyone but a career mainframe dev will ever pass, and they're not taking on anyone new.
Industry starts with the customers and the customers are trying to squeeze out costs as much as possible, shooting themselves in the foot.
Fun fact IBM pushes their hiring through a terrible 3rd party system that barely functions and doesn't even include options for common related degrees. IBM is incompetent and only exist because of wealthy corporatations paying politicians to have long intellectual property lifetimes.
I emailed their hiring service about it once and their response was essentially "take it up with someone else."
^ IBM is incompetent and only exist because of wealthy corporatations paying politicians to have long intellectual property lifetimes.
Not quite the case. Current mainframe users haven't migrated from it largely because the simplicity of scaling up, which means they can always buy a newer and more powerful machine when current one can no longer cope with the workload. I would expect it's much cheaper than migrate to other platforms in most if not all cases, considering the costs to develop and test the applications.
They can get away with it as long as they can find someone to carry out the upgrade/maintenance work. But if IBM keeps footgunnig itself, it won't belong before those users have to jump the ship as no one can work on it any more.
I didn't even think about it. I do get stumped on some convoluted headlines, but not this one.
The website The Register has its roots in England. My impression is that English newspapers have their own style and idioms for headlines that takes some getting used to. But once you do, the headlines become easy to read.
And, also, at the end, the article mentions that Fujitsu is finishing with mainframes in 10 years.
Young people are smarter than going aboard a sinking ship.