IBM swings shrink ray from workforce to mainframes

IBM has shrunk its z-series mainframes to fit 19-inch racks, the size used by just about every other server in the world. The new z14 range includes the ZR1, a single-rack machine said to improve on the smallest Z13 by improving throughput by 10 per cent while doubling memory to eight terabytes. The ten-core CPU hits 5.2Ghz – …

[GruntyMcPugh]

Fun 19" rack fact,....

... when I worked for IBM, down at one of their major datacentres (OK, it was Warwick), it had been deemed too expensive to buy IBM 19" racks (I know) so they went with some 3rd party racks, which were black, and looked the part. Except they weren't quite matched to the X Series, so if you needed to slide one of those puppies out, to say, add an extra processor,... to my surprise, you first had to remove the cabinet door.

So while I'm sure this bit of kit fits in a 19" rack space... it's probably a good idea to buy the IBM rack.

[Anonymous Coward]

Re: Fun 19" rack fact,....

> you first had to remove the cabinet door

Hmmm. Maybe that explains why there are some rack doors & sides stacked up in the junk area near my cube, labeled "IBM".

[Anonymous Coward]

Re: Fun 19" rack fact,....

Do they still use them?

[GruntyMcPugh]

Re: Fun 19" rack fact,....

"Do they still use them?"

No idea,... I changed jobs, and stopped doing hardware, now I don't work for them any longer (who does?), so haven't been down to that DC in ages. I'd like to think they've invested in new hardware, virtualised everything, and have it running on clustered, hyper-converged systems now. Mind you, I left in 2015, and we hadn't deployed any Server 2012 at that point,.....

[Anonymous Coward]

Re: Fun 19" rack fact,....

The racks in our server room have had all their sides and doors removed years ago, but my boss refuses to let me throw them away. Instead I had some shelving units in need of shelves and put the two together...

[Voland's right hand]

It is not a standard 19 inch rack

Looking at the picture it is double the max depth used in a normal datacenter.

So you still need to provision special space for it - same as a lot of other high end gear which is "sort-a" compliant to the 19 inch standard.

[GruntyMcPugh]

Re: It is not a standard 19 inch rack

Ah, so it is,....well, that would just be awkward,.... we'd have to put it into our cubes sideways if we wanted one.

[Doctor Syntax]

"Yet if the machines are as advanced and secure and generally wonderful as IBM insists, surely more buyers would already find them compelling?"

Price? Worry that as the vendor keeps shrinking there may be nobody left to support the kit?

[Ian Michael Gumby]

@Doctor Syntax ...

IBM has kept the mainframe as its sacred cow.

In the converged data center... the mainframe running linux , docker , kubernetes... could be an AWS killer if they get their pricing model down and develop similar tools.

That could be difficult unless Trump's Administration considers Amazon a monopoly and break ups the company.

[Anonymous Coward]

Re: @Doctor Syntax ...

How could they consider AWS a monopoly and not Microsoft? They'd have to breakup both and I think that'd be wonderful.

[Ian Michael Gumby]

@AC ... Re: @Doctor Syntax ...

Being a monopoly is a legal decision. They almost did consider it a monopoly when it came to the browser, however Microsoft consented and avoided that label.

That said. Amazon is a monopoly if the courts find it to be. And that's an interesting pickle. Some of their business practices on the retail side are in fact monopolistic. So IMHO it could be easier to show Amazon as a monopoly and then force the company to split up. That said... look what happened when they split up Ma Bell in to Baby bells. Some good and some bad. And somethings didn't change.

[Christian Berger]

I'd like to have an honest non-marketing answer to the question...

... what's so special about those boxes?

I mean you can run Docker images on cheaper hardware, too.

[Anonymous Coward]

Re: I'd like to have an honest non-marketing answer to the question...

Easy - keep on running those legacy applications...

[Ian Michael Gumby]

@ Christian Berger Re: I'd like to have an honest non-marketing answer to the question...

Cheaper hardware?

That's relative. Considering that your cheaper hardware is going to be less efficient meaning you're going to have to buy more hardware, larger footprint and more energy.

So it becomes a bit of a toss up and TCO is a bit fungible when you consider all of the issues.

FD, I haven't run all of the numbers so its still a question of which wins out. Also ... IBM's price is fungible. Large players get larger discounts. Tiers from A-J and then some discounts on top of that too.

[Ken 16]

Re: I'd like to have an honest non-marketing answer to the question...

It is incredibly hard to find non-marketing answers to this. IBM and CA have one set of answers, everyone selling x86 have another. Mainframe techies love them.

My own answers based on reading around are;

1) Reliable - they are designed to just keep ticking

2) IOPS - they handle really high volumes

3) Security - new-ish but they do incorporate in at fundamental levels

4) Financial arithmetic - they are optimised to add decimals (not simply integers) correctly

Everything else relies on believing someone about cost/performance and, since IBM can dial that up and down at will for every contract, I'll leave it out.

I'd like to hear a good answer to why Linux on z rather than on another platform?

[returnofthemus]

Why Linux on z rather than on another platform?

Considering that Linux was introduced on Z back in 1999 that question appears a little late in the day, especially as the total number of MIPS that run Linux has been growing steadily at a 45% CAGR.

However, I imagine the question you're probably asking is why would it be considered today, for which I can give you a very simple analogy....

Is not the Range Rover the best commercially available off-road vehicle in the world?

[Ken 16]

Re: Why Linux on z rather than on another platform?

My analogy is different > the mainframe is a shed. If you have a shed and you have some junk around the house then you may decide to box them up and move them out into the shed. Would you go buy a shed just to host your junk or would you leave it on the shelves in different rooms?

If you don't already have a mainframe and have a linux workload you can run distributed, what is the use case for purchasing a mainframe? If you have both distributed and mainframe, what is the driver for boxing up the workloads and moving them to z?

[returnofthemus]

Re: Why Linux on z rather than on another platform?

>>My analogy is different > the mainframe is a shed.<<

PMSL!

Where I live, the local authority still provide us a weekly refuse collection, all my junk goes in the bin.

However, I see you too are confused!

In this instance you are not going to buy a mainframe to distribute your linux workloads, you're going to buy one to CONSOLIDATE them!

A simple rule of thumb; 'Consolidate where you can, Distribute where you must'.

PS I voted you up for the comical analogy.

[Ken 16]

terminology

'Distributed' is my shorthand for x86 commodity servers - which provide the hosts within a virtualisation cluster so really any move to mainframe is putting VMs into LPARs (or are IBM calling them VMs too now) and I was using 'Junk' as a collective term for any useless household ornaments which have to be kept for sentimental reasons and not just thrown out.

I can see some efficiency benefits in theory but at the price of vendor lock in and difficulty in finding skilled staff.

[returnofthemus]

Re: terminology - I was using 'Junk' as a collective term....

Yes, it was rather crasp and somewhat illogical, I'm not sure anyone in their right mind would go out and buy a 'shed', ....potentially costing more than a house...., to store sentimental household ornaments.

Having never worked on a mainframe and far from being a mainframe expert, I'm struggling with your 'shorthand, my understanding is that you get a choice of either LPAR, zVM or KVM, which can also co-exist side-by-side on the same platform.

When did Vendor lock-in become a dysphemism for consolidating high-volume OLTP and OLAP mission critical linux workloads on the most reliable, most secure and most performant commercially available platform in the world?

Astonishingly, it's the one platform the majority of technical folk don't quite get, what leading business folk do, which is probably the reason why it has been around for 50-years+.

[Anonymous Coward]

Re: I'd like to have an honest non-marketing answer to the question...

I'd like to hear a good answer to why Linux on z rather than on another platform?

Because it will keep me in a job for another couple years? Oh, sorry, you meant how it benefits *you*. Sorry, can't help you there.

[Anonymous Coward]

That's a personalised version of the IBM management answer isn't it?

Because it will keep people investing in mainframe until their stock options vest.

[Michael Wojcik]

Re: I'd like to have an honest non-marketing answer to the question...

I'd like to hear a good answer to why Linux on z rather than on another platform?

Sometimes simply because the organization already had a big investment in z and was testing the waters for migrating some of the application load to Linux. Often the applications in question were written in Java, so moving them was relatively easy.

When zLinux first came out, virtualization was also a big selling point. The z hypervisor (basically a stripped-down version of IBM's VM OS, the granddaddy of hypervisors) could accommodate a huge number of tenants - I recall Usenet posts describing tens of thousands of zLinux LPARs on a sysplex. There are use cases for that sort of thing, and doing it with, say, VMWare on x86 hardware around the turn of the century would have been a hassle, particularly to manage.

These days with containerization (and particularly the rapid growth in management and orchestration tooling) the virtualization features zLlinux are no doubt less of an advantage.

But I still see a fair number of organizations investing in zLinux, and they have some reason for doing so. We've even had customers put our mainframe-environment emulation systems on zLinux, running CICS, IMS, and batch workloads under zLinux rather than in conventional zOS LPARs.

[returnofthemus]

Re: I'd like to have an honest non-marketing answer to the question...

I guess it's a matter of how many docker images you require and the nature of your application.

Bare in mind unlike x86 hardware, you don't get Spectre & Meltdown built-in ;-)

[Wazzupp]

Re: I'd like to have an honest non-marketing answer to the question...

Never underestimate the value of reliable infrastructure!!! I work for a company that manages several environments. I work on a team managing x86/Linux, there is alot more work involved in maintaining that environment's resiliency than a similar # of workloads on a mainframe.

Over the years additional resilience capabilites have become available for cheaper/non proprietary hardware, ofcourse you can run Docker on them too, you just have to spend more time planning for the underlying infrastructure anytime a change has to be made

[Hans Neeson-Bumpsadese]

When I first started in this business I remember a mainframe as being classified as a computer the size of a room, a minicomputer as something the size of a large cupboard or two, and a micro being something small enough to fit on desk. Seeing a mainframe shrunk to the size of a single rack unit makes me realise how far technology has come, and makes me feel rather old.

[Frederic Bloggs]

Mainframe? Are you sure?

To me, probably being of a certain age, the key distinction has always been: how much data and in parallel, can one shove through the computer at a time. "Lots continuously" = "mainframe", "quite a bit, but rather bursty" = "mini computer", "user interaction limited" = "PC".

So the designation "mainframe" has always hinged on the number and bandwidth of connections between the CPU and the storage devices and other streaming peripherals. While processors have got faster, I don't think either the interconnects or the hardware that handles them inside the "mainframe" have kept up to the same extent. Indeed the tendency to land stuff on buses, as opposed to direct connects closer to the CPU(s), will always add contention and slow dataflows down.

A "mainframe", then, is optimised to get the largest number of "streams" of data into, munged a bit, and then out of the unit as possible. Just because CPUs, discs and their interfaces are faster, doesn't alter that basic distinction.

I wonder just how far IBM can go before someone notices that they are just buying a (very) large and expensive mini?

[Anonymous Coward]

I want to know if they intentionally used Chevrolet's trademark.

[Flakk]

They May Actually Sell a Few More...

...if they included a Corvette with the purchase of every mainframe.

[Anonymous Coward]

How long before they bring out a Z/28?

[Anonymous Coward]

It's not exactly a new thing...

Not exactly news, given that the 9370 came out in 1986. I'd imagine that this one is a bit beefier throughput-wise though.

Possibly good news for those of us for whom a zPDT is a bit wee - a z12 BC is a distinctly chunky machine in comparison.

[jelabarre59]

I'm a mainframe

"You been runnin' your program on your little PC

Tryin' to fix your business with your 1-2-3;

That little Junior jus' can't manage your files

Well, I got some hardware that will make you smile

Chorus: 'Cause I'm a mainframe, baby

I got a hard disk, too

An' I can do twenty times

What that mini floppy driver can do"

<br>

https://genius.com/Bob-franke-im-a-mainframe-baby-lyrics

(and how do you get this site to properly format lyrics?)

[mako23]

Legacy Apps

The only reason to buy a mainframe is legacy apps. However the hardware is not special and neither is the operating system. Linux can run for weeks without booting. My mobile phone running android does its all the time.

I grew up with mainframes and my first job was a large mainframe site. I then moved onto as400’s a machine I still love. However I cannot see new customers moving onto a mainframe. With ultra fast SSD disks on a PC server the advantage is not there anymore. I doubt IBM mainframe business will die but it will become a smaller and smaller business.

Eventually IBM will leave this business, either by selling out or reduced investment.