IBM's z10 mainframe to take on the upstart PC

Shocking as it may seem, mainframe computing has never really gone away -- even in this age of modular PS3-based supercomputers, financial institutions, retailers, and other large corporations still buy the big iron, which means IBM still makes it. The company's latest, the fridge-sized System z10, follows up on the million-dollar System z9 released three years ago with faster, cooler processors, more energy efficient designs, 70 percent more computing capacity -- and a smaller price tag, starting at just south of a million dollars this time. Hilariously, the z10 caused a bit of a mainframe Osbourne effect: eager customers holding off on z9 purchases in anticipation of the z10 caused a 15 percent drop in IBM's mainframe revenue last quarter. Mainframe fanboys? Nothing surprises us anymore.

[bob sakamano @ Feb 26th 2008 3:40PM]

looks pretty

[Bryan Thornsberry @ Feb 26th 2008 3:42PM]

Somewhere there's a mainframe blog.

Somewhere within the comment section, somebody probably asked the question... "But will it run MS Calculator?"

[Carbonize @ Feb 26th 2008 3:47PM]

Pah I was actually going to ask if it would run Crysis at full size :|

[Liam @ Feb 26th 2008 3:49PM]

There's a blog for everything. This may or may not be a good thing.

[Tom Chandler @ Feb 27th 2008 1:19PM]

Gah, I was going to ask about Crysis too. Still anyone ready for an upgrade to give us the answer?

[ethana2 @ Feb 26th 2008 9:11PM]

Qalculate is to MS calculator what Ubuntu is to Windows 2000.

No one asked that question.

[I LOVE THE CAPS LOCK KEY @ Feb 29th 2008 1:47AM]

Yes it will run thousands of simultanious incarnations of MS CALC.EXE but no matter what, it still lacks the power to run Norton Antivirus. There not one computer around that can run that crap and still have free resources available for other programs.

[satanhimself @ Feb 26th 2008 3:45PM]

Hey, that mainframe certainly looks good. Maybe IBM will be releasing a console based on it?

AA
http://satanhimself.wordpress.com/

[phanbouy @ Feb 26th 2008 3:48PM]

Cool the new XBox!

[mike @ Feb 26th 2008 4:11PM]

Nope, that the thinhg is much to small to be the Xbox!

[mattydread @ Feb 26th 2008 4:15PM]

Indeed. That's probably just the power brick pictured.

[ethana2 @ Feb 26th 2008 9:12PM]

That thing lets you run your own code. The 360 is but a toy. The ps3 is not, and as of a few days ago, the Wii is not. Homebrew is crucial for getting real work done.

[Abuzar @ Feb 26th 2008 11:15PM]

Hmm, and I have most of my fun on that toy. I do MY real work on a REAL computer.

[Beau @ Feb 26th 2008 3:51PM]

"and much smaller price tag, starting at just south of a million dollars this time."

Um... read the post on the Z9 a little more carefully. It's R&D was $1B, but the cost was about a million. So basically this is slightly cheaper with 70% more performance, not even Moore would be too pleased.

[OneLove @ Feb 26th 2008 3:59PM]

will it play call of duty 4?

[Rocketboy @ Feb 26th 2008 4:01PM]

" Hilariously, the z10 caused a bit of a mainfraime Osbourne effect: eager customers holding off on z9 purchases in anticipation of the z10 " Why is that funny? That people are holding off buying something, because a better version of it is due to be released?

[murray @ Feb 26th 2008 5:30PM]

I think Nilay was just amused that stuff like that also happens in the world of million dollar mainframes, as opposed to thousand dollar PCs.

[packetsniffer @ Feb 26th 2008 5:52PM]

Then "amusingly" would have been a better word :-)

[Brad @ Feb 27th 2008 4:36AM]

What, that purchasing managers hold off on multi-million dollar purchases knowing that a better, cheaper product is around the corner? I don't see the hilarity...

[Tony Rayo @ Mar 1st 2008 5:03PM]

The same thing happened to the Dreamcast. It was released, there were some good titles to play too, but I'd guess based on polls and message boards at the time, the Dreamcast lost close to 80% of "potential customers" because the PS2 was on the horizon. In the end I'm kind of glad though, even though it hurt Sega's (or should I say Sega) pocketbook, they have released many good (and some not so good) games on a mutli-platform basis. The 2004 joint venture of Sega and Sammy (famous makers of pachinko machines if you've never been to Japan... btw US arcades have NOTHING on Japanese arcades when it comes to noise levels), Sega Sammy Holdings (which includes Sega of America) has really saved the Sega name and I hope to see Sega continue on as a software/arcade juggernaut.

BTW, I wish I had the funds and the need for a mainframe system like this (although personally I like IBM's hot-swapable blade systems better and HAVE actually got to use those in real life during a summer job as a web designer/makeshift sysop).

[felix @ Feb 26th 2008 4:02PM]

Is it really surprising that people would wait a quarter to get a device that's cheaper, faster and more energy efficient? Really?

[Vern @ Feb 26th 2008 4:11PM]

will it fit in a manila envelope? wakka wakka.

[Mark @ Feb 26th 2008 4:36PM]

No, but it will fit in Manilla.

[Jesse S @ Feb 26th 2008 4:11PM]

So, what do mainframes do these days? Just curious.

[jmiday @ Feb 26th 2008 4:25PM]

folding (away from home)

[Karl @ Feb 26th 2008 4:27PM]

The Telco I use to work for used a zSeries for customer billing and order processing.

[satanhimself @ Feb 26th 2008 4:36PM]

"Pah I was actually going to ask if it would run Crysis at full size :|"

No, but the mainframe they build after it just might:)

AA.
http://satanhimself.wordpress.com/

[theflew @ Feb 26th 2008 6:24PM]

Ever wonder how credit card processing, bank transactions, flight scheduling happen - That would be a mainframe doing the work.

[Matt @ Feb 27th 2008 1:33AM]

jmiday: Mainframes aren't supercomputers.

Mainframes are designed to deal with processing large amounts of data like banking and billing information quickly and efficiently. Supercomputers are designed to use a massive amount of power focused on solving one single question.

[Peter K. @ Feb 28th 2008 12:08AM]

they run things like large and critical DB2 installs .. for banks mostly.

[monkfishbandana @ Feb 26th 2008 4:14PM]

Whoa whoa whoa! Why has no-one notified me of these machines before? If I'd known, I'd have gotten one sooner!

Seriously, your friend may have a top-spec gaming PC with case lights and 40 fans - but there is no WAY that this can be trumped.

[Carbonize @ Feb 26th 2008 5:28PM]

Dunno I think this could come close - http://www.engadget.com/2008/02/24/exascale-computing-its-the-new-terascale/

Now not only will that play Crysis at full screen (possibly) it will also count the number of Apple stories on Engadget and plan the extermination of mankind at the same time.

Not sure which of these would take the most processing power.

[mattydread @ Feb 26th 2008 4:17PM]

Science!

...and very large databases.

[cmdwedge @ Feb 26th 2008 4:23PM]

I work on a site with a lot of these IBM z9's, and let me tell you - there's not a cooler looking server (rack) around. It really looks the business. Of course, I look after the Windows servers.. so I don't get to play with the Mainframe machines. Those guys are WEIRD. :)

[Joshua Ochs @ Feb 26th 2008 4:25PM]

Mainframes still command loyalty because of one simple fact - they never go down.

NEVER.

Screw weeks or months of uptime, these have years and decades. Including OS or hardware upgrades, which can be done without taking the system offline. If you need processing that simply does not stop, you get a mainframe, simple as that.

Read up: http://en.wikipedia.org/wiki/Mainframe_computer

[AlexL @ Feb 26th 2008 4:57PM]

How is this different from distributed computing where many tightly-coupled or loosely-coupled computing elements work together to perform tasks? One failed computing element has no impact on the uptime of the entire cluster.

[Matt @ Feb 26th 2008 5:27PM]

"How is this different from distributed computing where many tightly-coupled or loosely-coupled computing elements work together to perform tasks? One failed computing element has no impact on the uptime of the entire cluster."

Its different because its in one box.

Instead of having 1000-2000 x86 machines all taking up power, space and pumping out loads of heat and noise you have a single box running 1000-2000 guest OSs. Savings in electricity alone are thought to be about 85% which can be a significant amount of money (hundreds of thousands a year apparently), then you save loads of cash on support and maintenance...imagine troubleshooting a network problem in a server room with 2000 machines and the associated costs!

Then there are various other benefits, for example the internal "network" interconnects between the guest OSs (its really all done in software) are running significantly faster than ethernet can ever dream of (several gig/s), you can increase your processing capacity significantly in an instant ("On Demand" - call IBM and they remotely turn on more processors/memory/storage) instead of building, installing, testing, rolling out new hardware - a lengthy process that wont react quickly enough to a Slashdotting/Digging! Then you have centralised management, 24/7/365/100/∞ uptime, IBM support, backward compatibility to the 60s and beyond.

Its good stuff.

[AlexL @ Feb 26th 2008 5:56PM]

"Instead of having 1000-2000 x86 machines all taking up power, space and pumping out loads of heat and noise you have a single box running 1000-2000 guest OSs."

The boxes don't use power (although I agree about space consumption), it's the components inside the boxes. More redundant components mean more reliability, so if you cut down on redundancy you are cutting down on reliability. And nobody said anything about x86.

"the internal "network" interconnects between the guest OSs (its really all done in software) are running significantly faster than ethernet can ever dream of (several gig/s)"

10Gbit Ethernet, Infiniband, etc.

"you can increase your processing capacity significantly in an instant ("On Demand" - call IBM and they remotely turn on more processors/memory/storage) instead of building, installing, testing, rolling out new hardware - a lengthy process that wont react quickly enough to a Slashdotting/Digging!"

What are you saying, that by default the mainframe has some capacity disabled? How is this an advantage? You can disable components of any redundant system and claim you can now "remotely turn on" more capacity.

[Mace Moneta @ Feb 26th 2008 6:22PM]

"The boxes don't use power (although I agree about space consumption), it's the components inside the boxes. More redundant components mean more reliability, so if you cut down on redundancy you are cutting down on reliability. And nobody said anything about x86."

Mainframes don't use off-the-shelf components. The logic has multiple planes - the active plane, the diagnostic plane and the spare plane. The diagnostic plane continuously test gates on the active plane during the brief moments when they are idle. When a weak gate is detected, a spare gate is switched in dynamically. If too many spare gates are used, a service threshold is crossed. A service call is then initiated by the service console, and the part requiring replacement is specified. A technician shows up with the exact part required to perform the replacement - all while the system is in use. There is no higher level of computing reliability available.

The typically environmental savings by using a mainframe over discrete processing devices can easily amount to several hundred thousand a year (AC+electricity+floor space).

"10Gbit Ethernet, Infiniband, etc."

For mainframes, we are talking tens of gigaBYTES per second. Essentially, the virtual network interconnects are very high bandwidth and very low latency memory-to-memory moves. These operations can be taking place between hundreds or thousands of virtual guests concurrently.

"What are you saying, that by default the mainframe has some capacity disabled? How is this an advantage? You can disable components of any redundant system and claim you can now "remotely turn on" more capacity."

You pay for what you need. Rather than having to deal with physical hardware changes, unpaid for hardware can be dynamically actuated. You pay more, you get more - on the fly.

In addition, the virtual guests can be configured to dynamically share resources, allowing a "slashdotted" guest to dynamically increase its resources enormously. It like saying "Oh crap, traffic is spiking!" and running out on to the data center floor to connect 1000 additional server blades to the cluster. Except no one has to lift a finger to do anything.

[AlexL @ Feb 26th 2008 6:31PM]

"You pay for what you need. Rather than having to deal with physical hardware changes, unpaid for hardware can be dynamically actuated. You pay more, you get more - on the fly.

In addition, the virtual guests can be configured to dynamically share resources, allowing a "slashdotted" guest to dynamically increase its resources enormously. It like saying "Oh crap, traffic is spiking!" and running out on to the data center floor to connect 1000 additional server blades to the cluster. Except no one has to lift a finger to do anything."

Ok, I see the benefit of the rest of what you are saying. But I still don't see how this part is a mainframe-only feature. Cluster systems can just as easily start out with disabled components with additional components enabled on the fly should demand increase, and loads can be balanced dynamically between tasks as well.

[Mace Moneta @ Feb 26th 2008 6:53PM]

"Ok, I see the benefit of the rest of what you are saying. But I still don't see how this part is a mainframe-only feature. Cluster systems can just as easily start out with disabled components with additional components enabled on the fly should demand increase, and loads can be balanced dynamically between tasks as well."

Cluster systems could be done this way, and if you can point me to a cluster vendor that does it would be great. Essentially, I want to order a cluster with 250 nodes, and have the vendor deliver 1000 nodes, but still charge me for 250. You know, just in case I decide I need more. If I decide I want to go to 500 nodes, I just call the vendor and he remotely powers on another 250 for me, and ups my monthly billing. When I'm done with the 500, I call and tell them to drop be back to 300 nodes, and my billing does likewise.

That's what we're talking about with mainframe disabled capacity.

With regards to dynamic reconfigurability, think about having dozens of clusters, and spending all day swinging blades and chassis back and forth between clusters as load changes. The alternative is over engineering those clusters, and paying for capacity that only needed intermittently. Because all the capacity is in a single system image, the OS (z/VM in large guest environments) or firmware (PR/SM in small number of guest environments) can dynamically swing resources where needed, within limits defined by the system staff. You aggregate the total capacity requirement and then can swing it manually or automatically easier than entering a comment here.

Seriously, by comparison, dealing with blade chassis and multiple small (in comparison) clusters is hard work and expensive, once the scale starts to go up.

[Adamwho @ Feb 26th 2008 4:40PM]

Main Frame computing gone away?!!

Show me a PC where you can allocate up to 256G of RAM to invert a giant matrix!

[Superprime @ Feb 26th 2008 4:53PM]

So when do you know you need a mainframe?

[Yubastard @ Feb 26th 2008 9:01PM]

good question... I want one, but I don't know if it's right for me.

[iofthestorm @ Feb 26th 2008 10:12PM]

If you have to ask, you don't need one.

[murray @ Feb 26th 2008 5:33PM]

Man I would love to add this to my VMWare LabManager pool.

[Trey @ Feb 27th 2008 12:36AM]

Lab Manager only runs on x86 hardware.

[murray @ Feb 27th 2008 12:46AM]

It would be running ESX Server... but yeah, same story I guess. sigh.

[donjeske @ Feb 26th 2008 6:50PM]

"Instead of having 1000-2000 x86 machines all taking up power, space and pumping out loads of heat and noise you have a single box running 1000-2000 guest OSs."

'The boxes don't use power (although I agree about space consumption), it's the components inside the boxes. More redundant components mean more reliability, so if you cut down on redundancy you are cutting down on reliability. And nobody said anything about x86.'

The IBM press release mentioned the ability to virtualize 1500 x86 machines on a Z10. At least in the x86 environment the only redundant element is the power supply and those either load share or are active passive.

"the internal "network" interconnects between the guest OSs (its really all done in software) are running significantly faster than ethernet can ever dream of (several gig/s)"

'10Gbit Ethernet, Infiniband, etc.'

The Z9 did 10Gbit Ethernet and 4Gbit Fibrechannel. The Z10 adds Infiniband for clustering mainframes. Communications between hosts within the same Zseries are much quicker, though. If one guest requests data from another guest its a memory move at the speed of the backplane. There is no IO and no protocol overhead.

One thing that has not been mentioned is IBM's pricing for its 'specialty' engines. Say you bought a Z9 (or z990 or even older versions of the mainframe) and had loaded it up with the IFL engines that run Linux. You have bought those engines for life. If you upgrade from a Z9 to a Z10, those engines are free. Your Linux workload just got a 70% bump for free. You still pay for the chassis and your traditional workload engines (if any.)

'What are you saying, that by default the mainframe has some capacity disabled? How is this an advantage? You can disable components of any redundant system and claim you can now "remotely turn on" more capacity.'

There are no redundant features disabled, other than spare CPU's that get turned on for free, concurrently, if one of the other ones should fail. When IBM started selling CMOS based air-cooled mainframes, they would configure the box with at least one spare CPU for this purpose. IBM started marketing access to more of these additional engines that can be turned on concurrently as a feature. They can be turned on for spikes in load, as part of disaster recovery and for planned growth. Memory can be added concurrently in the same way. Once contracts are set up, users can access the additional capacity whenever they need it.

[Ray @ Feb 26th 2008 7:31PM]

Nothing compares to a mainframe. I am a Windows Admin and a Z/OS Mainframe systems programmer. I much rather deal with a mainframe. The security on a mainframe cannot be beat. Servers are hacked everyday.. Not to mention that you can set up Lpars and NALCS that will save you on
software costs.