
There's a good chance you think
mainframes are about as cool as pocket protectors, your parents on Facebook, and COBOL... the latter of which, of course, is largely executed on mainframes. If so, stand still while
IBM blows the doors (and the dust) off of your mainframe misconceptions with its new zEnterprise 196, offering 96 5.2GHz cores, 3TB of RAM, and hot swappable I/O drawers for when you need to change pants in a hurry. All this is said to boost performance by 60 percent compared to its predecessor, the
z10, while also reducing energy consumption by a claimed 80 percent -- though that could be compared to people sitting in tanning booths performing calculations with abacuses for all we know. However, you can drop consumption a further 12 percent by opting for the water cooling system, nice if your AC unit is already struggling. IBM will start shipping these behemoths sometime in the last quarter of this year and didn't mention pre-orders, so get ready to rent the biggest truck you can find and camp out in Armonk if you want yours on release day.
Show full PR text
New York, - 22 Jul 2010: -Industry first: IBM mainframe governance and management benefits extended to select POWER7, System x environments(1)
-New technology boosts complex analytics performance up to 10X,(2) delivering insight from business data in minutes versus hours
-zEnterprise is the most powerful, scalable mainframe server ever – up to 60% faster than System z10(3) for new workloads
-Most energy efficient mainframe--more than 100,000 virtualized servers can be managed as a single system; 60% more capacity than z10 while using same amount of energy
IBM (NYSE: IBM) today announced the zEnterprise mainframe server and a new systems design that allows workloads on mainframe, POWER7 and System x servers to share resources and be managed as a single, virtualized system. The new mainframe is also the most powerful and energy-efficient mainframe ever.
The new systems design combines IBM's new zEnterprise mainframe server with new technology--the IBM zEnterprise BladeCenter Extension and the IBM zEnterprise Unified Resource Manager--that enable it to manage workloads running across System z, and select POWER7 and System x servers. The new technology is the result of an investment of more than $1.5 billion in IBM research and development as well as more than three years of collaboration with some of IBM's top clients around the world.
As a result, customers can integrate the management of zEnterprise System resources as a single system and extend mainframe qualities, such as governance and manageability, to workloads running on select IBM POWER7 and System x blade servers. With the ability to manage workloads across systems as one, the zEnterprise System can drive up to 40% lower acquisition costs and reduce cost of ownership by 55%.(4)
IBM designed the zEnterprise System to address an important issue for corporate data centers -- the jumble of disparate technologies added over time to run specific applications and which operate in silos, sometimes unable to communicate with each other in real time and requiring separate staff and software tools to manage. This long-standing challenge for customers is aggravated by dramatic increases in cost and complexity amid a rising tide of sophisticated, data-intensive workloads in an increasingly interconnected world.
For example, using the zEnterprise System with the zEnterprise BladeCenter Extension and IBM zEnterprise Unified Resource Manager, a financial services company managing credit card transactions on the mainframe using an IBM blade optimized for analytics can gain insights from the information in seconds. Previously, it would have taken hours for the two disparate systems to integrate their databases. IBM estimates that complex database queries can experience up to a ten-fold performance improvement in this hybrid environment.(2) In addition, with IBM's new design, the financial services company can extend the mainframe's always-on reliable qualities to its customer service applications running on IBM blades servers.
At Citi, IBM System z plays a major role as a core processing engine of virtually every line of business at the global financial services company.
"The new IBM zEnterprise System represents a potentially revolutionary change to the platform and the next phase in the evolution of highly efficient, scalable processing opening up the possibility of hosting entire workloads on a single highly integrated system," said Martin Kennedy, Managing Director, Citi's Enterprise System Infrastructure. "The new zEnterprise also paves the way to enhance the energy dynamics of our data centers. As one of America's greenest banks we plan to take full advantage of the additional capacity and advanced power and cooling capabilities unique to zEnterprise. Citi's unified technology decision making model and its recent efforts to gain efficiencies prepared us to invest in these innovative technologies that benefit our clients."
"The new IBM zEnterprise System represents a bold move to fundamentally change how data centers are managed," said Tom Rosamilia, General Manager, IBM Power and z Systems. "The new mainframe is the fastest enterprise server in the world and represents a giant leap forward in performance. This new dimension in enterprise computing-- extending mainframe governance to POWER7 and System x blades integrated into the zEnterprise System architecture--was developed over the past three years with direct involvement from a team of IBM's 30 top customers, which provided direct input at every stage of the development process."
IBM zEnterprise BladeCenter Extension
The IBM zEnterprise BladeCenter Extension allows supports purpose IBM POWER7 and System x BladeCenter systems as well as blades optimized for specific workloads, such as analytics and managing Web infrastructure.
IBM blade servers inside the IBM zEnterprise BladeCenter Extension--which can be managed like mainframe resources--integrate with System z and can run tens of thousands of off-the-shelf applications. Later this year, IBM will deliver the zEnterprise BladeCenter Extension with support for IBM POWER7 blades running AIX, IBM's UNIX operating system. IBM is also introducing the IBM Smart Analytics Optimizer to accelerate the performance of complex analytic workloads at a lower cost per transaction.
Next year, IBM plans to announce additional general purpose blades for the IBM zEnterprise BladeCenter Extension including select IBM System x-based blades running Linux. Additional workload optimized blades are planned to include IBM DataPower for improving website and network performance.
I have 10 in my backyard.
@uckApple
How the hell do you cool 96 5.2GHz cores??
@tobsmonster2 Very carefully ;)
Can it play Crysis 2 with max AA?
@hero785
I don't think even Crysis could really challenge this type of machine. Instead what you could do is have your entire LAN party playing Crysis against each other in separate virtual machines. Now that's a serious gaming machine.
@ashwinkn
Wouldn't that be something!
@ashwinkn I just wet myself... o_O
@loocas LOL
@hero785
Actually it probably can't play Crysis well, I doubt it has much in the way of a video card :( All CPU, no GPU... Boourns
@hero785 Wernt the graphics in crysis 2 dumbed down for the consoles anyway?
@the bandit
:O
"nice if your AC unit is already struggling"
Still generate (and therefore must dissipate) the same amount of heat. Water cooling just makes it more efficient at getting the heat into the air.
@TMoney4Life But it also makes it easier to direct the heat, because there is just the radiator (mainly) dissipating the heat, instead of heat coming out of all the vents.
@Bobinator2000
Unless the mainframes radiator can be mounted outside or in another room, directing the radiator's heat does not matter. The rooms AC unit will still have to overcome the same amount of heat whether the mainframe is air or water cooled. Think about it....
@vmod32 The water-cooled models connect directly to the datacenter's chilled water loop. There is no radiator.
what do companies use these things for nowadays? as far as i know, rack servers can handle pretty much all corporate computing requirements. if it's for research, don't academics go for supercomputers instead?
@tonicboy Companies use mainframes where they need the most reliable and secure computations, e.g. financial and transactional number crunching on a massive scale. IBM has invented some novel ways of ensuring 99.999% up-time. With this model, IBM is also allowing intel/power/cell blades to be installed INSIDE the mainframe, so you can even get rid of your rack servers and run everything from a single environment.
@madduffy I like to add that Mainframe are generally use to do lots of IOPS through network (millions & millions).
And I will correct that in fact the intel/power/cell blades are not install INSIDE the mainframe but in a frame next to the mainframe.
@tonicboy - What madduffy said. Plus, the next time you go to an ATM, make an airline reservation, do practically anything thru a financial institution, thank a mainframe. ;-)
One cool thing you can do is boot up thousands of VM Linux instances. So, instead of having thousands of PC boxes sitting around, you can have thousands of users compiling their own kernels and booting them up.
This ain't no toy. ;-)
@tonicboy
We have 3 z9 business class mainframes, and we use them for revenue accounting, database ops, inline applications and backend computing. One of our frames carries over 9 billion instructions a second, running at bus speed with storage IOPS over fibre.
We just call them the huge, black mystery machine in the department of expensive things.
@Derek Hanson
I've been through a phase where I wanted to find out just *what* an IBM mainframe really is and I've come out convinced that today (i.e. not speaking 30-40 years ago) it's almost sure completely uncompetitive nowadays and is simply a way for IBM to extend their vendor lock-in to customers who bought one last century and are convinced they can't afford to migrate. Most of the basis for this is that IBM absolutely does not allow publishing *any and all* benchmarks between their mainframes and "normal" systems (what they call "distributed" - as in "non-centralized") but continue to cite ludicrous performance and energy savings numbers. I've tried both official (as a journalist for a local IT magazine) and unofficial channels.
Practically all IBM numbers regarding mainframes are bogus and taken out of context by IBM themselves. For instance, the "9 billion instructions a second" number quoted in the comment might seem big until you realize that mainframe CPUs are beefed-up PowerPC cores which are basically RISC, so they count each clock cycle as an "instruction". In other words, clock for clock, you get more performance out of a quad-core CPU which can be bought over the counter for about $1300 (e.g. Xeon 5570).
Reliability cannot possibly beat that achieved by distributed applications and distributed data centers like Google and Facebook have, simply going from first principles. Etc. Etc.
In short - I hope noone buys into IBM Mainframe stories without benchmarking (performance and money-wise!) everything for themselves first.
@ivoras Both your 'facts' and your conclusions are wrong. First, jumping to the conclusion that mainframes are 'uncompetitive' because they use different benchmarks is like saying locomotives are uncompetitive because they don't publish MPG and 0-60 numbers. The reason IBM does not use 'standard' benchmarks is because the machines don't run (and aren't designed to run) 'standard' workloads. Customers who run traditional mainframe workloads run mixes of CICS, IMS, and DB2 transactions. Some customers run all batch. Some run all on-line. Many run a mix. The machines are highly tunable to the customers needs. There are no 'standard' benchmarks that are useful for those workloads. Customers who run 'new' mainframe workloads don't run one copy of Linux, they run hundreds or thousands of them. There is no 'standard' benchmark for that either (well there is, but it was only released a week ago).
Next, the instructions that are counted are zSeries instructions, not clock cycles or Power (which has nothing to do with zSeries). The zSeries instruction set has several hundred opcodes (more than oe hundred were added in this release alone) and is most definately not a RISC processor.
Comparing Google or Facebook to a mainframe is absurd. Google and Facebook give the appearance of high reliability, but they are not necessarily all that reliable. Sometimes you do get errors (timeouts, incorrect results, etc) on Google or FB, so you just hit the reload button and go on your merry way. You would not be happy if your bank or credit card used the same techinques. In those environments EVERY transaction must work EVERY time, with no ambiguity.
Lastly, nobody buys a mainframe based on something they read somewhere. Instead, they take their actual workload (programs and data) to IBM and try it on real hardware.
@ivoras
IBM's mainframes will spank the hell out of any x84 based server farm in pure I/O performance. That is where relational DBMS's get their massive performance gain on the mainframes. And IBM's mainframes can actually achieve five nines reliability(5 minutes downtime per year), where even Google(who's server farms are the biggest of all) does not expect nor have that kind of reliability.
In any case there are new installations of mainframes all the time, because there are some huge advantages in certain cases, even if a lot of workload can be just as easily done on commodity H/W.
BTW, the processors in the mainframes are not even designed by the same department as Power processors.
PS: Even Google uses the traditional methods of accounting and payments. It's their search and less critical applications that you hear about.
PPS: You will not find any benchmarks of Oracle's DBMS anywhere also. It's a normal procedure.
BTW
These beasts are $110,999
When I saw the picture I first thought "Those must be the new XBOX 360 Slim's everyone's going on about"
@bigwhitebear18
The design is actually very cool, and that green line down the middle does groove with XBox styling.
...and indeed it is slimmer than the XBox 360.
@bigwhitebear18
Probably is quieter and generates less heat than a 360.
I, for one, think mainframes have always been hot and are even cool today.
Wow, imagine how fast those cards would fly at you after you won in Solitaire! I might have to get one of these things.
@the bandit
Exactly what kind of giant desk do you have?
3TB of RAM, that's just absurd. D:
@Slygathor Well, you could use the last model to replace 5000 x86/linux servers. I think that number just went up...
@Slygathor Some folks need the ability to join 3-4 10GB database tables. It helps if you can keep a few gig from each table in RAM.
@fais Turn it sideways and it becomes a pretty rad, liquid-cooled desk, itself.
3TB of RAM. I NERDGASMD
Does it blend? More importantly, do they make blenders that big?
@Kelmon Will it bend? No, probably not. Will it shred is more appropriate here... http://www.youtube.com/watch#!v=j1_uvM-5xKs&feature=related
One side of it is a vendor machine
mainframes were cool all the time ;)
@uckApple
I don't know where you got that. It cost hundreds of thousands of dollars per core.
We've got one of these zEnterprise badboys for our back-end where I work... would make a great home server too :D !!
"White models of IBM's new zEnterprise 196 have continued to be more challenging to manufacture than we originally expected, and as a result they will not be available until later this year. The availability of the more popular zEnterprise 196 black models is not affected."
@ThisWas
they better make sure these things dont need a friggin bumper or a case
@xtian63
That was quite funny actually.
Then again if something goes wrong with those toys no bumper would help ;)
I will take 2 plz!
@uckApple
More like a cool mil, easy. These are the high class enterprise ES frames. The smaller, business class z11's come out next year.
Actually they ship out of Poughkeepsie not Armonk
Facebook will never the same using this...
With three of these scattered in different I/O centers across the country, you could build one hardcore cloud framework.
Now if I only had a few million bucks to play with . . .
"96 5.2GHz cores"
One (geek) has to wonder : What's that chip ?