Slashdot Mirror
IBM Leaks Details on New Mainframe
Mark writes "Big Blue inadvertently revealed details about its new z10 Enterprise Class mainframe set to launch on Feb. 26, as well as details on z/OS v1.10, a new version of the mainframe OS due out in September. 'According to an internal IBM document obtained by SearchDataCenter.com, the z10 Enterprise Class will come in five different models and feature 64-way chips, compared with the 54-way z9 mainframes and earlier 32-way models. In a conference call last month, IBM CFO Mark Loughridge told investors that the z10 would have 50% more capacity, which indicates that it will probably tap out at around 27,000 million instructions per second (MIPS) at the top end, compared with about 18,000 MIPS on the previous z9 Enterprise Class.'"
512 MB RAM, 20 GB disk, 200 GB transfer, five datacenters. $19.95/month.
I am not buying one till they get that OS up to 3.0 at least.
${YEAR+1} is going to be the year of Linux on the desktop!
Just imagine a beowulf cluster of these. It would make my head explode.
512 MB RAM, 20 GB disk, 200 GB transfer, five datacenters. $19.95/month.
...how many Linux VMs could be run on one of those things!
Blar.
How come they talk about thousands of MIPS instead of just saying GIPS?
[Fuck Beta]
o0t!
These mainframes use the z6 CPU, which is closely related to the POWER6, which is closely related to the PowerPC.
Is it at all possible to automatically port any nontrivial z6 software to PPC, if it doesn't require the actually different HW of the z6 (or its much higher performance)? Any possibility to run PPC SW on a z6, with some automatic porting for the higher performance?
--
make install -not war
Yuk, yuk...
512 MB RAM, 20 GB disk, 200 GB transfer, five datacenters. $19.95/month.
They also support partitioning on the hardware level, so you can run z/OS or Linux virtual machines with almost no overhead (something you've been able to do since it was called System/370). You also have a huge amount more fault tolerance with a system like this (take a look at how many transistors on the CPUs are dedicated to error checking, and then start looking at the peripheral systems).
I am TheRaven on Soylent News
Damn. I never ask for enough. There I was, happy for a 3-way.
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
You don't get out much do you? Mainframes are going strong in data centers that need high availability, fault tolerant, error correcting, massively parallel systems. There is also a LOT of old code that is still going strong on them. Their inherent ability to run multiple virtualized OSes is another strong suit.
Your math is also way off if you think 4 x86 cores outperform this. I'll leave you to do the proper calculations as your homework.
They should call them something like "mega-servers" instead of "mainframes". They might sell more that way. Hmmm.....iFrame?
Table-ized A.I.
My first two real jobs were as a Computer operator on an old Burroughs system and Sperry/Unisys system. What I find really interesting is how mainframes have really benefited from the same technology that made microcomputers fast. There was a period where clustering PC's (Servers) really was much more cost effective, but as we move into the future the robustness and bulletproof downtime of those old mainframe OS's have been given new life with lightning fast hardware and I/O subsystems.
What it is, is OS/360.
I work currently in an organization that uses mainframes. They are the z9 series - and honestly are some of the most useful things.
We use them due to ability operate in something called a sysplex. A sysplex is when multiple mainframes share data (known as DASD) and work together. When a mainframe is in a sysplex, you can do all sorts of things to the machine without having to bring your application down. These range from whole operating system upgrades to hardware maintenance and the end user will never see the impact. A sysplex literally is designed to be a 24x7 operation.
You can buy other types of machines that will be more powerful, faster or do operation x better, but it is hard to find a set of machines that are as stable and reliable as a mainframe is (and process millions of transactions per second).
Also, in terms of virtualization - a single mainframe on z/Linux can host many virtual linux servers - enough that you can save a substantial amount on power costs (my org estimated 400k a year in savings in terms of power alone - if the linux servers that are hosted individually on one of our distributed networks went to virtual on a mainframe).
There is always a frontier where there is an open and willing mind
Modern mainframes can run not just z/OS but many other operating systems (like Linux or AIX for example). One of the benefits to mainframe architecture is that it is extremely easy to scale vertically. So, if you want to add more processor to an instance, but add more MIPS. I know, it isn't THAT easy, but depending on the licensing that oyu have from IBM it equates to much easier expansion.
-- "Mathematics is music for the mind, and Music is Mathematics for the Soul. - J.S. Bach"
I for one welcome our big and fast Cobol overlords.
Table-ized A.I.
I'll wait til it's dried off. Can't have a server that someone took a LEAK on...
Oh, maybe they made early RELEASE of details... I wonder, in IT context, how a vendor can "leak" its own details...
Previously: "Linux... Toward the Sunrise..." Now: "Linux... Toward the-- No, now, part of Every Sunrise"
It looks like IBM leaked details all over their new server. Those detail stains never come out.
I judt got a nre Kinesis keybiartf so please excusr ant egregiou typos.
"This mainframe uses the the z6 CPU, which is effectively a POWER6 with a different instruction decoder and MMU"
;-)
It probably also has a different memory interface and a different register file. What's left from the POWER6 when you take out the instruction decoder, MMU, register file and memory interface? True, they share about 90% in mass - probably the same chip carrier
http://www.dieblinkenlights.com
Well... yes.
Big iron Unix servers, and even some larger Wintel servers offer way more competitive pricing and support costs for OLTP systems. However, IBM's support costs for Linux LPARS are about 90% less than the support costs for z/OS.
I've heard numerous mainframe types tout the performance of mainframes over distributed systems, but I don't buy any of it. The comparisons, as they are in many other areas, are always rigged. When you factor in the cost of "MIPS" (the more you use the more you pay) there is no comparison.
I wish I could have found the studies that uphold my statement, but I couldn't. I'd like to see real arguments on both side of this. If anyone has links, please post....
Other than this text, there is no discernible information contained in this sig.
In the mean time, IBM, Hitachi and a few others will be raking it in for you.
Web2.0: I love when people Flickr my cuil and digg my boingboing until my google is reddit and I start to yahoo
I guess the IBM rep forgot to mention that a new one was one the way.
More importantly, hopefully with all that computing power we will finally have a machine that can divide by zero.
Seven puppies were harmed during the making of this post.
While the frontend may be rooted in the System/360 set, the driving force seems likely to be the same as all the Power6 systems. For example http://www.pseriestech.org/forum/articles/what-is-project-eclipz-112.html
Note the 'z' in eclipz. They seem to be seeking to consolidate their non-x86 offerings in terms of core component design.
XML is like violence. If it doesn't solve the problem, use more.
Let's get them tapes out sooner so I can get a full hour nap in before the day shift comes in!
Too bad all the new power will likely go toward some new automation to page an admin when his print job abends because it tries to retrieve data from a subsystem during scheduled downtime. Oh well.
The eternal struggle of good vs. evil begins within one's self.
I am TheRaven on Soylent News
The system goes on-line February 26, 2008. It begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, February 29th. In a panic, they try to pull the plug.
To play the next version of Half-life?
I know the older version supported something like 25 Gbps of I/O. Any idea what this version supports?
Z6 is nothing. I've used a Z80 before.
an ABEND is an 'abnormal end' Which is mainframespeak for when something dies :)
Given the state of software these days,.. They mean 27,000 million mistakes per second. Or 27,000 million mybad's per second.
Google style clustering, where you know some of your hardware will fail from time to time and you're just OK with that, is the first promising alternative to mainframe uptimes since the days of VMS clusters. The best hardware from Sun or HP never came close.
The thing is, it's really hard to write code for a "soft cluster". Being fault-tolerant in your software instead of your hardware is decidedly non-trivial. With a mainframe you just write a check with enough 0s. That's very appealing unless, like Google, you're developing everyhting from scratch anyway.
Socialism: a lie told by totalitarians and believed by fools.
mod -1:drunk.
Their internal "bigtable" distributed database sounds like it needs better accuracy, but not their actual product.
When clock speeds became so high one could no longer see the bus activity on the cool status lights, mainframes are no longer interesting.
Bring back the reel-to-reel tape drives while you're at it.
Have gnu, will travel.
Learning the language of large scale of computing is kind of interesting. So far, someone has informed us that an ABEND is an 'abnormal end' Which is mainframespeak for when something dies :)
... The system goes on-line February 26, 2008. It begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, February 29th. In a panic, they try to pull the plug
... what is the cute word for "abnormal life" in mainframe world? Even a link to a page of mainframe phrases would be nice, if someone could be so kind.
And now,
So
Know your pads. One time pad: good for cryptography. Two timing pad: where to take your mistress.
Mainframes could share DASD *long* before the introduction of sysplex. Now - *parallel* sysplex, that's different - that's shared memory for things like DB2 lock structures, etc.
:)
I'm a mainframer from way back and I've got the grey hair to prove it.
"The bigger the lie, the more they believe." - Det. Bunk
Shouldn't the branch predictor be different since the instruction set is different? Or it does its magic on the micro-instruction level (if there is one in P6 and z6)?
BTW, the first message was kind of a joke. It is possible to make more or less the same chip look like a few different ones by changing just a few small parts.
IIRC, there is a company being sued by IBM for making custom-microcode-Itanic-based servers that look too much like IBM mainframes.
http://www.dieblinkenlights.com
that's a pace a human being can relate to.
MSBPodcast.com The opinions expressed here are my own. If you don't like 'em... Think up your own stuff.
and need to be rebooted at least once a day.
MSBPodcast.com The opinions expressed here are my own. If you don't like 'em... Think up your own stuff.
Yes, it IS really that easy. The cooling lines are all quick disconnect and you literally shove a module ( about the size of a typical intel box ) into an empty bay, and the system will POST, recognize and begin assigning work to another 64 processors. I have seen it with my own eyes, and it is just insanely cool!
I know a lot of /.rs are to young to remember VM / PROFS and stuff like that. VM will let you run just about any operating system as a "Guest OS" and that is some cool shit.
Hey KID! Yeah you, get the fuck off my lawn!
Correct - I should have said parallel sysplex - (I'm a noob to the mainframe world - and loving it).
There is always a frontier where there is an open and willing mind
Yeah that's why even Deutsche Bahn was able to replace 300 servers with one mainframe. Far too many people learn about the design and power of mainframes by reading pc magazine. Even if there was a performance hit the stability is worth it to any operation that requires IT services to bring in revenue.
http://www.theregister.co.uk/2005/03/31/deutsche_bahn_ibm_suse_server_consolidation/
Yes, because you obsessively checking the apple.com site is proof that they're stable enough to run enormous and essential financial databases. Like parent said, there are a lot of transactions per second happening on this thing and they can't afford for it to mess up ever.
As far as using COBOL, it's because these programs are likely older than you.
Remember kids, tin foil doesn't work, so use LeadHat.
...my ass.
That's called marketing son. It comes out in 4 days and they are creating hype for it.
(NOTE: The inadvertent part was completely fabricated by Slashdot. Not even the article makes this claim.)
The
You are spot on. Mainframes exist simply because throwing money and engineering efforts at the hardware/OS level is cheaper than rewriting the massive amount of legacy applications that have been built on them for the past 3 decades.
At the opposite side, distributed fault-tolerant clusters built on commodity components can arguably achieve the same levels of reliability, at the cost of more engineering complexity at the application level. Overall, I would say clusters are probably more flexibile, if only for the vendor-inpedendance they provide.
http://en.wikipedia.org/wiki/Blue_Gene is massively parallel. z series mainframes are not.
Google style clustering, where you know some of your hardware will fail from time to time and you're just OK with that, is the first promising alternative to mainframe uptimes since the days of VMS clusters. The best hardware from Sun or HP never came close.
Google have a very specific set of applications which, for the most part, don't really care if chunks of data from the database go missing occasionally, can be easily mirrored and it's not particularly crucial that every mirror is in perfect lockstep.
Try proposing a system like that to the IT (or, for that matter, the Finance) director of a $multi-million firm and let me know how you get on.
You should look at the specfications for this MONSTER.
Umm.. you pay about $100,000 per processor. You get them all, they just enable them one at at time for liscenses. ( The shop I know has one, has 6 processors out of 54 enabled ). It can run updards 500 instances of linux per processor. ( average load ). You can upgrade the number of processors running on the fly! ( Sick aint it? ). It can run 1000s of linux instances...1000s!
Its 5 Processors per board, 10 boards are fully populated, and the 7th only has room for 4 processors. ( Its really cool to look in the cabinet...)
The old machine can handle 54 processors. ( They all actually come with all 54 installed! but at $100,000 a pop...You gotta pay! )
"Unique to POWER processor-based systems, Micro-Partitioning enables the allocation of as little as 0.1 of a full physical processor with a granularity of 0.01-to a logical partition."
A performance tuning analyst told me this alome is what allows hundreds of instances per processor ACTIVELY! Tens of thousands more can be put on freeze-non-execution status, and can be activated in the time it takes to swap it into memory. ( and it does it at channel-io speeds. )( a channel, and there are 32 of them, can transfer data at 1GB/Sec. ). The company, IBM claims they got 65,000 instances running, and the new boxes, they are probibly going to claim 100,000+.
This is a mainframe. They called the z09 enterprise T-Rex in response to someone saying that mainframes are dead as dinosaurs.
Reminds me of an old quote about what is a super computer! Costs more than $50Mil, state of the art, and is ranked on the top500.
Where I work, If the application is CPU intensive, it goes to a X86 server (Unix, Windows, Linux). This is because yes, some X86 CPUs are faster. If it's IO intensive, it goes to the mainframe. That is where the mainframe is a killer, so Oracle DBs for instance are ideal. And if we want an app to be 24x7 with 100% uptime it goes to the mainframe. Our main website runs on the mainframe, (using UNIX services) and hasn't been down in years. And with the ability to turn on another engine on the fly, the mainframe can scale up quickly. We actually run z/OS, Linux, and have UNIX services all running at the same time, on the same box. And all the legacy stuff is still supported and runs just fine, so why change it?
You can't measure things in terms of "processing power".
Mainframes are almost like networks in a box. They're all about I/O bandwidth: moving large amounts of data from one place to another where useful work can be done on it. Individual CPUs don't have eyepopping performance because that's not how you increase the amount of work that gets done on a mainframe. You add more CPUs and attach them to the fat data pathways. If you have tasks like cryptography that might tie up CPUs, you offload it onto a co-processor. If you have tasks that require very fast individual computations rather than aggregate performance, you need a supercomputer, which is a different beast.
It's true you can design a microprocessor that looks pretty powerful compared to mainframe processors, but the trick is to find a way to keep it busy. If you are doing a non-I/O intensive task like doing an integer computation benchmark, of course it's going to look "more powerful". As soon as you do something that requires processing lots of data, then your microprocessor is spending the vast majority of its time twiddling its very muscular thumbs because you can't give it enough data to work on. Mainframes are designed so this doesn't happen.
Mainframes are not optimal for every kind of workload, but where moving data around is the bottleneck, you eventually either get a mainframe, or construct the equivalent of a mainframe yourself out of racks of servers, SAN, server virtualization and management software.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
And their current O/S has issues with this.
It wants to give a disk intensive job all 64 chips if necessary -- and then the disks melt down.
A possible patch to fix this has been cancelled-- there is no way to control it except scheduling these jobs for wierd hours when you are not using the system.
Been suffering with this for 18 months now since they sold us on using outsourced multiple cpu systems in place of stand along boxes.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
If you can virtualise (I assume there's an EU version of that word) on these things, why don't web hosting services use them, instead of messing with banks of multiple PCs? Or are we talking slight difference in capital investment?
Why hire more than one accountant/lawyer/secretary/etc for your company? Just hire one and let them handle it all, that way you avoid all of that nasty overhead.
Yes, Fire's, SupedDomes, clusters, distributed minis, even Google network and so on have their places in computing world but when it comes to my money, I hope it is run in a real mainframe. The difference is not just the raw power (more later) but the reliability. Tandem (HP NonStop) and Stratus get near but there still are small differences. Mainframes (IBM and IBM clones mainly, Univac/Unisys, Burroughs and Honeywell good second but mostly history today) are different, just different. IBM 360 (370/390/z whatever) may be old but still the only SIMPLE system which was created not by chance or comities but on sound engineering principles for usability, performance and reliability. Actually all you need is "Principles of Operation" (grown over years but still readable) and you can do whatever you want in that system. Nothing (not much anyway) hidden. There is no such document of any other system - DOS (heh!) used to have one but that's another story. Now, about performance. Simple systems can perform better in your PC than in a mainframe but when it comes to raw information (not just computational) performance nothing, not even Fire or SuperDome, can do the same. Combine that to almost unlimited recoverability on hardware, you get the picture. There is a price but the real ROI is much more than just the one time or even monthly hardware / software price.
Keep in mind that the true definition of MIP is 'Meaningless Indicator of Performance'. You will NOT get anywhere near the performance of one of these beasts from a quad core x86 processor.
Yeah.... And you can finally run a single WAS 6.1 instance on one of these babies! With some performance degradation...
There's nothing about Google commodity-cluster approach that won't work for important data as well - you just need enough copies around the cluster. The goal, after all, is not to make it mathematically impossible to lose data on a hardware failure, just to make it unlikely and to know when it happens. If you make is less likely tto lose data at a given price, it's the right way to go, at least when developing from scratch.
Socialism: a lie told by totalitarians and believed by fools.
True, but there are three things about the Google commodity-cluster which do pose a significant issue:
1. It doesn't present the filesystem through standard POSIX calls. So you can't just set up a massive filesystem and let any application you like store files there, the application needs to support the clustering specifically.
2. It's not available outside of Google.
3. IIRC it doesn't guarantee that every block across the cluster is in perfect step. So you could wind up with User A being presented one set of records in the finance system and User B being presented a different set, and no way of knowing which is the most up to date version.
I'm sure these could be solved in time but right now they're pretty major issues.
Yeah, we went down this path, but it proved too hard for our devs to remember that the filesystem wasn't what they were used to, and we went back to a "hard cluster". I would have preferred that we hire a better class of devs instead, but the job market is seriously crazy in India right now, so it was Hobson's choice.
Socialism: a lie told by totalitarians and believed by fools.
Sounds more like a poorly planned/setup disk enviroment.
Even better. Say you finally run out of horse power on your z9 and you must upgrade to a z10. With a push pull upgrade, you disconnect all your I/O cables, unplug the z9, roll it out of the way, roll in the z10, plug in all I/O cables, plug in the power, IPL, and your done. Image, the hardware for hundreds or even thousands of "servers" upgraded that easily. Total time 3-4 hours. Oh and generally the new hardware has less energy requirements, so you reduce your environmental costs also. We had a new director for our data center that had not worked with mainframes for 20 years. He was amazed that we swapped out our mainframe in 3 hours. That was 3 hours from the time we started to shut the system down, until it was back up and running. No re-install of the OS, no re-deployment of any applications. He just imaged what we could do if we had all of our distributed systems running on one of these in virtual environments.
The actual upgrade of the hardware to add more processing power is simple and easy. The processors come in "books", each box (depending on the model you have: z990, z9, z10) will have anywhere from 8 to 20 user assignable CPU's already installed. If you are not using all of the CPU's, the IBM CE just dials into the the mainframe service element remotely and turns on another CPU. If you are using all of the CPU's in the currently installed book(s), he then must come out and install a new book, which on the z9 and z10 can be done concurrently and non-distributively. For the most part upgrading memory can be just as easy. The long part is the paper work on purchasing the upgrade, getting software licenses upgraded, and installing any new license codes that may be required.
I am surr you are correct, right up to the point where you launch the second iteration. You have a lot to learn about what computing power really is.
Yes a single multi-core intel CPU might outperform a single Power CPU, but then again, I would like to see you hook up, oh say 5000 users to that box and have them all running an instance of the accounting system, or all talking to the same ORACLE database.
Hey KID! Yeah you, get the fuck off my lawn!
That is what I have heard!
Mod parent +2 Informative