Having worked in BigCo land for 20 years, your estimates are way off.

There are F500s that have 1k mainframes each, today, in 2022.

Thank you for setting that straight. Ther more I read through his answer, the more ludicrous it became.

There should be a way to flag and remove comments that don't violate any guidelines, but are just objectively wrong, in a way that doesn't penalize the author. Like "you tried, but nothing about this is correct, so it goes to the bottom of the comment section above all the dead/shadowbanned comments."

I prefer the back-and-forth the way it is now. I perversely enjoy reading a comment and "buying into it" only to realize from a reply that I was naive or too trusting. Or, at least, that there are more angles to the story.

This helps you build up an immune system that is skeptical. If most wrong answers were removed/flagged/hidden then the remaining wrong but unflagged would be even more a problem.

Exactly. This helps train the "don't believe everything you read" muscle.

Sure for the percentage of people who do read all the way through and then rethink things. This is <1% in most communities

Assuming everyone is an idiot, and designing around that basis, seems like a great way to foster a community of idiots.

And unlike stackoverflow, there’s not really any good reason for anyone to think a top-level comment is somehow authoritative, especially because the best comments on HN tend to be replies

> I perversely enjoy reading a comment and "buying into it" only to realize from a reply that I was naive or too trusting.

Good for debate and free speech in general, but bad for your grasp of the facts in the case every time there don't happen to be any corrective replies.

Hard to weighs pros vs cons.

(ObSillyPun: Naah, just put them on the scales! But remember that a lighter pro will often beat a heavier con anyway!)

Is reply and downvote not enough?

I want to avoid the downvotes, though. Downvoting is for things that are off topic or don't add to the conversation, or comments made in bad faith. Someone who just happens to be wrong may very well be commenting earnestly in good faith.

Your bullshit detector should be trained continuously, not put to sleep by others labeling a post true or not. A dangerous path.

So you DO search for answers to mainframe questions on s**overflow?

there are many medium size banks that have a mainframe or two at the heart of their system.

Even small banks. Basically every bank that was around 20 or 30 years ago started with mainframes (or very large minis) and still uses them. I'm not in that business, but I've seen the server rooms, and I haven't heard of a bank switching over to a rack of PCs, yet. There might be, but that would still hundreds, perhaps a thousand banks in Europe alone running on large-ish iron.

There might be less mainframes in total, though. The bank where my parents worked had 4 at different sites, but now runs on a single one, AFAIK.

Unisys ClearPath Dorado is just a fancy Xeon machine running an emulator.

But the OS, database, network definition and workflow languages are still proprietary. Which goes back to the original question: where is that knowledge store/discussed?

Inside the company. Not gonna be a youtube tutorial lmao

"Sup nerds, today we're going to be writing some COBOL, but first don't forget to hit the bell and smash that subscribe button..."

Not even kidding, I'd probably watch some live-streaming of that kind of thing.

Twitch for mainframes... COBitch!

> I haven't heard of a bank switching over to a rack of PCs, yet.

Right. The main limitation is the core software platform. This is an extraordinarily complex system that requires expensive regulatory sign-off/audit/etc.

There are some vendors in the space who are trying to do a new bank + new core. Even if someone had an awesome x86 clustered core system with 100% regulatory approval, any existing mid-tier bank converting to it would be an ~8 figure ordeal.

It seems like their point still stands. If there were 100 million mainframes, x86 would still have more, and the standards for skills and experience would still be different due to the nature of deployment.

> So, that 80-90% of the Fortune 500 is really just 1000 mainframes total. Generously, if every Fortune 500 had a pair in each geography (roughly EU, Americas, Asia for big corps), that ends up being 6, so 3000 globally, in production.

As others have posted, your estimates are WILDLY off.

First, fortune 500 companies dont own "two" mainframes, but a LOT more.

Next, many non-F500 companies own a lot of them as well.

Years ago I worked for a local cable company in Canada.

Not a "F500" company, owned at least two mainframes for the "head office" and each different 'subdivision' owned their own mainframe.

> Really just 1000 mainframes total.

I'd be willing to bet that any major city has more within its borders.

I don't know if iSeries (AS/400) counts as Mainframe... But the software stack is similar and I know that hundreds (if not thousands) of casinos run old Main Frame era software.

You can add just about every auto parts retail store to that list.

Lots of community banks & credit unions run on top of iSeries right now too.

AS/400 was retired in 2013

And they're minicomputers, not mainframes :)

The software stack is very similar...

Even if OP's estimates are "WILDLY off", there still aren't many out there

Multiply his estimate to 1000 per F500 company ... that's only half a million

Put a pair in every major geography ... that's only 5,000,000

And, for kicks and giggles, multiple by another 10 for funzies

That 50,000,000

Yes, it's "a lot" ... but it's nothing compared to commodity hardware

According to a quick web search, Google alone had in excess of 1,000,000 x86 servers about a decade ago

That's one (large) company

I've worked with myriad companies who're running 10s or 100s of 1000s of x86 servers (let alone x86 desktops/workstations/laptops) - with nary a mainframe in sight (or maybe a couple ... which is on the order of "none" next to 100s of 1000s or millions of x86 boxes)

Per Microsoft [0], there are 1.4 billion Windows 10 and 11 devices (yes, I know that's not a server count, but it speaks to the orders of magnitude difference between x86 devices and mainframes ...even with my generous estimates posted above)

-----------

[0] https://news.microsoft.com/bythenumbers/en/windowsdevices

The flip side of that is that commodity hardware runs commodity software. It’s entirely possible to run a very large company without having any bespoke software.

Mainframes on the other hand or more likely than not to have software that only runs on a handful of machines at most.

So number of machines is not a good proxy for amount of software written.

On the other hand I’ve never touched a mainframe so it’s all speculation…

>It’s entirely possible to run a very large company without having any bespoke software.

I find that claim remarkably improbable - the overwhelming majority of all code written is for in-house business tasks

>So number of machines is not a good proxy for amount of software written.

It is a good proxy for the relative need of developers, however - with billions of x86 devices out there vs max a couple-few million mainframes...the scale of need is just wildly different

I can confirm after working with the IBM mainframe people a bit. The internal training of these systems make SO/ random other information sources on the subject pointless.