Running a full IBM System/370 Mainframe on a Raspberry Pi Zero | Hacker News Reader

Running a full IBM System/370 Mainframe on a Raspberry Pi Zero

twitter.com

165 pointslhoff50 comments

This is apparently greatly over-hyped (or a complete lie, depending on your interpretation) :

https://mobile.twitter.com/mainframed767/status/125801882682...

It's possible the original tweet poster was running a much newer mainframe OS on his Hercules emulator, but didn't want to say it directly, since he might get a license bill for that.

"7 times faster than System/370" is an odd claim though. There's several models and 19 years of upgrades.

System/370 ended in the eighties I think? Maybe it is reasonable to be 7 times faster than the fastest System/370.

The fastest S/370 (pre-S/390) machine in the line was, I think, the 3090; these were multiprocessors, but a single processor had a 69 MhZ clock rate. Even if you assume it retired one instruction per cycle (and I'm not at all sure it did), it wouldn't be a shock for Hercules on an RPi to manage that.

Aside: “Soldier of Fortran” is a great user ID!

And the login screen is just taken from the manual for the MVS TK-4 system: https://mobile.twitter.com/mainframed767/status/125802863861...

So the pictures are: a stock photo of a mainframe, a stock photo of a Raspberry Pi, and a cropped screenshot from an MVS installation manual. Looking at OP's other tweets he appears to be a pretty shamelessly self-promoting "thought leader" type so I would be pretty unsurprised if it's bullshit.

The counter tweet isn't particularly levelheaded either. 5 years to compile something on a platform that has a four 1.5Ghz cores? Listing a bunch of software that the original code may or may not need?

It's been a while since I went to Twitter, though. Honestly HTML 1.0 with blink and construction gifs has less information pollution, once all the bad memes, youtube channel spam, and bad one-line retweets spill onto the screen.

> four 1.5Ghz cores

The Pi Zero is actually only a single-core 1GHz board with 512mb RAM. It's fairly slow compared to the rest of the lineup. I couldn't even upgrade my Pi 2 to a Zero in a few of my projects.

This guy had a reputation when I frequented Quora of being a bit full of shit, but convincing.

Should probably be a link to the Hercules emulator, since it's the real magic here. http://www.hercules-390.eu/

Even runs current z/OS, though there is no practical legal way to do that.

http://wotho.ethz.ch/tk4-/ just in case

How is it relates to Hercules?

Tk4- is a custom "Hercules ready" packaging of the last version of MVS (3.8j, 24 bit, released in 1981) that you're legally allowed to use with Hercules. And probably other nice add-ons around installation, docs, etc.

http://www.bsp-gmbh.com/turnkey/

"The MVS Tur(n)key system is a package of freely available parts of the MVS 3.8J Operating System. The Tur(n)key System does not (to the best of my knowledge) contain any copyrighted or licenced software by IBM or any other company. Especially, the MVS 3.8 Operating System is not OS/390, nor z/OS, or any other such thing. The provided hardware emulator (Hercules) may perfectly run these operating systems, but you would need to obtain them, and a licence, yourself.

Also keep in mind, please, that CICS, RACF, DB2, ISPF etc are all licenced program products and as such are not available in the Turnkey distribution."

It uses Hercules emulator:

https://en.wikipedia.org/wiki/Hercules_(emulator)

See there:

"Operating systems which may legally be run, without license costs, on Hercules"

MVS is:

https://en.wikipedia.org/wiki/MVS

"Multiple Virtual Storage"

"Developer IBM"

"Initial release 1974"

"MVS releases up to 3.8j (24-bit, released in 1981) were freely available and it is now possible to run the MVS 3.8j release in mainframe emulators for free.[2]"

"MVS originally supported 24-bit addressing (i.e., up to 16 MB). As the underlying hardware progressed, it supported 31-bit (XA and ESA; up to 2048 MB) and now (as z/OS) 64-bit addressing. The most significant motives for the rapid upgrade to 31-bit addressing were the growth of large transaction-processing networks, mostly controlled by CICS, which ran in a single address space—and the DB2 relational database management system needed more than 8 MB of application address space to run efficiently."

"The system is typically used in business and banking, and applications are often written in COBOL. COBOL programs were traditionally used with transaction processing systems like IMS and CICS. For a program running in CICS, special EXEC CICS statements are inserted in the COBOL source code. A preprocessor (translator) replaces those EXEC CICS statements with the appropriate COBOL code to call CICS before the program is compiled — not altogether unlike SQL used to call DB2."

"MVS/ESA: MVS Enterprise System Architecture. Version of MVS, first introduced as MVS/SP Version 3 in February 1988. Replaced by/renamed as OS/390 late 1995 and subsequently as z/OS."

That is necessary information to understand the comment:

https://mobile.twitter.com/mainframed767/status/125801882682...

"Bull. Shit. Theres no TCP stack. Theres no cics, no DB2, no ims."

What's certain is that, using just that, one can't run all the "real life" sources that typically depend on CICS or DB2.

But also see icedchai's message: https://news.ycombinator.com/item?id=23091687

Also, if somebody would think it's easy to hack that using more common tools and on more common platforms:

"The native character encoding scheme of MVS and its peripherals is EBCDIC" and not ASCII.

I had z/OS running on mine using a torrent. Unfortunately I had no idea how to use it and lost interest. Definitely not the same thing, but I find DEC OSes (VMS, especially) much more accessible.

Hercules is a great emulator. I used it to run z/Linux (debian, IIRC) some years ago when I needed to do a port of some libraries and had no access to the real hardware.

IBM, of course, very much wish that Hercules didn't exist. I remember them stamping with great prejudice on a company, some time in the mid 00s, who were offering a sort of System Z overflow capacity box. It was basically Hercules running on a beefy x86_64 box, and it was being sold as a way to do more System Z stuff without buying more capacity from big blue.

I find it somewhat ironic that IBM is struggling so hard to find COBOL programmers to keep mainframe alive, but they also insist on locking down z/os to the point that you can't get access to a terminal unless you already know COBOL and work for an enterprise spending millions with IBM.

You want nerds to flock to your platform, you have to give them access...

There is the z/PDT, which is an emulator that runs on your PC. It comes with a software pack that includes z/OS, z/VM, and few other things. It runs $900 a year, not sure if there is an initial cost.

http://dtsc.dfw.ibm.com/adcd/adcd.shtml

However 900 a year is still out of reach for most individuals. They should make it free.

Has anyone done this? Is it $900 for everything one needs or are there additional costs?

I think the $900 is just for the z software. For zPDT you need to provide your own hardware/cloud instance, and pay for the emulation software on top of that. That came in around $2.5k/year IIRC.

It is just the software. You get the software and a USB dongle that acts as the license key for the emulator.

I used this back when I worked at IBM. Most people just ran it on a ThinkPad running Linux. I ran it on a rather beefy tower system, as it was slow on the ThinkPad.

We called the tower system, the baby z, and I took it and physically placed it inside of our System z. One day our CE asked why there was a tower inside of the System z. I told him that it was pregnant.

> I find it somewhat ironic that IBM is struggling so hard to find cheap COBOL programmers to keep mainframe alive

Fixed that for you. There are lots and lots of former Cobol developers that have been laid off over the past 20 years due to offshoring. There is no shortage.

There's no problem finding "cobol programmers"; "cobol programming ability" isn't their bottleneck. Finding someone competent to do professional software archaeology & borderline forensics for 40K/y is their bottleneck.

“Shortage of programmers” almost always means “we don’t want to pay full price for programmers”.

Isn't that the definition of shortage? When we say there is a shortage of X, it means that there is X, but not enough of it to keep the prices at an affordable level, so the prices have risen and some people who need X, want X, and/or previously could afford X no longer can. This is true as long as there is a shortage, not an absolute lack of something.

I mean, famines happened because sometimes there would be a shortage of food, perhaps due to drought, or a sudden increase in population not matched by an increase in food production. Prices would go through the roof. Rich people could afford the new prices and persevered. Poor people died of hunger. But if the poor mustered enough gold, they could go buy food like rich people did.

So my question is, do you consider famines shortages? If not, what even is a shortage? In yes, how is labour shortage any different, except in scale?

The mere fact of something being expensive doesn't mean that there is a shortage of it. A shortage is when there is less of a given thing than there is demand for it. Typically, demand goes down when price goes up and there is no longer a shortage, so a lack of essential goods like water would be considered a shortage because demand cannot retreat in the way that demand for programmers can retreat.

Note that your example of a famine brings up 'shortage' before any mention of prices. There are multiple ideas of shortage at play. At the most basic level shortage can mean that there is not enough of a good/service for everyone to have it at any price (food in a famine, medicine in a pandemic). Typically we apply shortage to situations where the supply and demand curves reach some hard stop at the edges (elasticity goes to 0). In the case of normal market dynamics where supply and demand are both reasonably elastic then there really are no shortages per say - just cheapskates.

Finally! A comedian!

Perfectly setting up the classical joke that banks are too poor to find COBOL programmers because there is a "shortage" ;) ;)

You, sir, are a true connoisseur and made my day! Thank you!

> keep the prices at an affordable level

They're a bank. They can afford it.

They weren’t struggling to find COBOL programmers before this. There’s a pool of contractors to maintain COBOL systems. The problem is that there weren’t enough of them available when suddenly a ton of COBOL systems needed changes at once.

I can't believe this (still?) isn't being more widely publicized by IBM - they offer free 1-year z/OS accounts to anyone. You're perfectly free to re-apply the following year (I've done it myself).

https://www-01.ibm.com/events/wwe/ast/mtm/audit.nsf/enrollal...

I used to experiment with a Hercules MVS (my first real computer job was sysop at an institutional datacenter with an MVS) on my eeepc on my second tour in Iraq. Doing convoy security, of course we never had any network connection of any kind (it being the middle of the desert and also having active RF-jammers on our guntrucks), and there was always a long wait for the MWR computers at whatever base we'd stopped at for the night, so we were left with only whatever data we could pack before each mission. I spent hours and hours hacking on that thing. It was a pretty absurd situation, but I had a lot of fun.

On my last two sub deployments I bought a midrange laptop and Mandrake Linux 8.0, had a grand time installing, reinstalling, tweaking the hell out of it, learning to handle a running *nix.

Came in handy when an onboard system based on HP-UX cut out while underway and out of contact with fleet technical support.

I was on a surface (aka "target" to you) cruise, and spent all my downtime in the medical department hacking away on an unused little laptop. This was when MARSGRAMs were the height of fast communications, and the procedure for sending them was to handwrite your message on a slip of paper and pass it to the radio guys, who would then type a bunch of them into a computer hooked up to a packet modem. That was pretty low priority for them, understandably, so things would wait in queue for quite a while before they could get around to doing some data entry.

I knocked up a little QBasic app that let you type in your MARSGRAM message and then write it out to a floppy, batched alongside everyone else's messages. Once every couple of days, we'd run that floppy to the radio guys who'd pop it in their computer, blast out the messages, and write any replies they'd received to it back out so you could take them back to the department and everyone could read notes from their loved ones.

I'd have a hard time exaggerating how well this went over with everyone involved (except maybe the MARSGRAM operators who were suddenly carrying a lot more messages than they were before because it was so easy to send them). "Hey Captain, I heard you say you were missing your wife. Want to send her a message and hear a response in a couple of days instead of months?"

It worked out pretty well for me personally, too. I had a great boss at the time who told me flat-out that I was an idiot for pretending to want to work in medicine, when my heart was clearly elsewhere. It'd never occurred to me to make a career out of computer stuff before then, and a short time later I was out of the Navy and enrolled in a compsci program. Thanks for the kick in the right direction, Ken!

In 2006 on my first mission with MSF, we had a BGAN satellite terminal. Luxury! Because in my downtime, I'd read the tech library on CDROM, which explained how to organise a mail service with floppy disks and Landcruisers, with a relay to Paris over HF radio via Portishead, UK..

That's beautiful.

This is the power of Moore's law. A computer that cost millions reduced to $5. Shame we waste all the power we have now days.

The bitcoins must be mined!

For a bit more fun with mainframe emulation if you're not already a mainframe programmer, check out MUSIC/SP on the System390 emulator [1]. You can run MUSIC/SP on Hercules also, with some limitations [2].

[1] http://www.canpub.com/teammpg/de/sim390/

[2] http://www.canpub.com/teammpg/de/sim390/hercnote.txt

EDSAC[1], on BCPL[2], on raspberry Pi et al, ..

[1] https://www.cl.cam.ac.uk/~mr10/Edsac.html

[2] https://www.cl.cam.ac.uk/~mr10/BCPL.html

I wonder how the uptime and stability is on the Pi? IBM mainframes rarely go down and are up from setup to decommissioning.

IO bandwidth is also a serious issue. You can run 250K interactive users and 1K printers and 25 or so tape drives quite easily off a mainframe cluster. Probably not off a pi, LOL.

Except when they do go down. Not as a result of the processor itself, but as a result of the hiccups in the dependency chain.

In one case ours went down for a day when a backhoe hired for construction cut a power supply line. The data center batteries designed to give time for the desiel generator to kick-in were fine. The generator failed to start as a result of contaminated fuel despite monthly tests.

The second time was when an employee brought their 12 year old son into the raised floor environment. He found and pressed the haylon dump button for fire suppression. Staff were lucky to escape without suffocation.

The employee was fired and stricter entry requirements to the raised floor area were implemented.

> The generator failed to start as a result of contaminated fuel despite monthly tests.

I don’t think this is he first time I’ve heard such a thing. I wonder if the people running such tests know how long you have to run the generator to flush the fuel lines? And aren’t there failure modes for ICEs - and diesel in particular - where the engine stops working once it reaches operating temperatures?

I had a car that would occasionally conk out due to a damaged O2 sensor (due in turn to a crack in the manifold).

Raspis are cheap and simple. Fewer things to fail, but each part is more likely to fail. They have low cooling and power requirements though. I'd expect the storage to be the thing that fails most often.

The interesting experiment with raspis used in high uptime scenarios is when you make a cluster of them. If you aren't reliant on any one to stay up, then you can afford many more hardware failures than traditional clusters, albeit much less dense. You trade size, performance, and ease of maintenance for systemic reliability. Unfortunately, I can't find the articles that look at this.

IBM System/370 = $2,248,550 [0] Raspberry Pi Zero = $5 [1]

[0] https://www.ibm.com/ibm/history/exhibits/mainframe/mainframe... [1] https://www.adafruit.com/category/934?src=raspberrypi

That’s $2,248,550 in 1970 dollars. In today’s money, that would be more like $15 million.

The raspberry pi's cost increases when you buy the power supply, SD card, keyboard, display, and cables as well.

alternately, its much < $87k based on Computers, peripherals, and smart home assistants CPI-U.

https://www.bls.gov/cpi/tables/supplemental-files/home.htm

In the meantime, somewhere in the universe, people are buying mainframes ... for personal use: https://blog.mainframe.dev/2019/05/buying-ibm-mainframe.html. Go figure! :-)

I have a friend who used to work at IBM... and does Model railroading... there could be an interesting intersection there. 8)

Reply on news.ycombinator.com