ZAZ: I am surprised to read that NASA has no mainframes. What do they use when they want a reliable computer? Where do they keep their business records? Sure there are alternatives, but IBM is still big in big organizations.

IBM is mostly big now as a consulting arm, not in terms of mainframes.

Most mainframes now are just legacy systems. Organisations like banks don't use them for their reliability but because of the cost of conversion. A cluster of servers is more reliable than a mainframe - if you get a hardware problem then you can still continue processing.

Eddie Adams from Torrance: //JOB1 GOODBYE
//BYENOW EXEC PGM=GDBYWRLD
//SYSOUT DD SYSOUT=*

FTFY
(names limited to 8 characters and underscores are invalid. SYSIN as you had it was invalid. We'll ignore the jobcard)

There would still be mainframes if SUN, IBM, and HP weren't total douchebags about the support contracts. Or if they were relevant anymore. We still have two SUN machines (M8000's, I think - barely a mainframe, but SUN treats it like one). They cost over $200K a year just for support. We have one crazy old HP that still costs over 100 grand for support, and we barely use it anymore.

Compare that to any cluster you could possibly configure: Windows or linux (RH or SLES) or even Solaris x86 running in conjunction with a free hypervisor or VMware. That will give you comparable performance, redundancy out the ass, and you can run the whole thing on hardware so cheap it doesn't matter if you lose half your cluster nodes every year - you still save tons of money.

When we turned off our last IBM they came in and tried to sell us another machine for $750K for the hardware alone, plus a support contract well north of $100K. We told them we had already set up a SLES cluster to replace the old IBM box and everybody was happy with it. Support for a SLES cluster? $20K a year for premium support. Support for hardware? We didn't care, since we had cobbled the whole thing together with old Dell 2650's out of warranty. We eventually upgraded that whole cluster using Dell blades, and even THEN the hardware cost for a chassis and six blades was less than a fifth what the IBM quote was.

We're standing up a grid supercomputer this year for our research wing, and I'm excited about it because it's going to prove to our senior management once and for all that you don't have to spend umpteen million dollars to get a decent number crunching platform. These people still think you need a Cray to go balls out.

Tr0mBoNe:

ZAZ: That triggered a memory of the 1970s TV show "Salvage 1" where our junkyard astronaut heroes hack into a NASA mainframe to do some trajectory calculations, and the mainframe is shut down (or they are cut off) during a critical moment.

I am surprised to read that NASA has no mainframes. What do they use when they want a reliable computer? Where do they keep their business records? Sure there are alternatives, but IBM is still big in big organizations.

Clusters do the job of Mainframes but use Linux or Windows and normal engineers and administrators know how to use them. A couple racks full of blade centres, a large storage array (with a well thought out backup policy) and a big network will do the job. IBM sells all of that stuff. For several millions you can get the performance of a mainframe that cost several hundreds of millions.

What he said... I can't speak for NASA, but having done some work for the DOE you can sort of walk through their data centers and see the progression from big heavy iron being phased out to 7U rack systems to blade servers running cluster services.

With each step you're doing a lot better performance-wise.

ZAZ:

Reliable systems take a lot of software effort. I used to work on cluster software (c. 1999) and the test/debug time was much, much longer than planners had anticipated. There is a general habit in software development to skip error testing, but error handling is where you squeeze out a few more 9s of reliability. (We didn't have any 9s before I left that job.)

Given the speed of current networks, there's a lot of brute-force dumb things you can do for reliability. You don't have to care if node X goes down if it's marked bad and node Y has the same information.

So much for my AS/400 knowledge getting me a job at NASA.

I like that the article tried to explain what a mainframe is, or was. That's cute. Maybe it should have included a discussion about how email used to take a long time to arrive, if it arrived at all. Or perhaps how "online computing" used to be these bulletin board lists that were operated from people's houses and we would dial in, provide an access code, then be able to post something in writing that other members could read, and they could post something. A full page of text may take a week to develop. The bulletin board systems had cool names like "The Dungeon," had a hundred or so subscribers, and displayed blocky computer text. How wild was that?

Daisy, Daisy, give me your answer do.

/DNRTFA

Alright! Cape Cod has become a unit of measure! If I knew hexadecimal math I could tell you the equivalent in Rhode Islands.

Don't they know just how dangerous pulling the plug can be?


/hot like the M5

I'm not saying the article is wrong, yes a zSeries is a frame element no doubt. I'm not totally convinced the picture shown is of a 12 year old chassis though.

I'd also wonder what they're replacing it with, if the 'rack upon rack of Dells' I'm sure this modern thinking (i.e. could find neither arse nor elbow with a map, torch and labels stabled to each) CIO has undoubtedly plumped for is actually cheaper than the $30kp/a power bill when looked at system capacity. Yes POWER is quite energy hungry, so are legions of Xeon's.

Tom_Slick:

So much for my AS/400 knowledge getting me a job at NASA.

United Parcel Service. Just saying.

Wow. I did not know that.

And had no idea that servers had replaced mainframes.

HAL? We hardely knew ye...

ZAZ: That triggered a memory of the 1970s TV show "Salvage 1" where our junkyard astronaut heroes hack into a NASA mainframe to do some trajectory calculations, and the mainframe is shut down (or they are cut off) during a critical moment.

I am surprised to read that NASA has no mainframes. What do they use when they want a reliable computer? Where do they keep their business records? Sure there are alternatives, but IBM is still big in big organizations.

Client server baby.

farkeruk: ZAZ: I am surprised to read that NASA has no mainframes. What do they use when they want a reliable computer? Where do they keep their business records? Sure there are alternatives, but IBM is still big in big organizations.

IBM is mostly big now as a consulting arm, not in terms of mainframes.

Most mainframes now are just legacy systems. Organisations like banks don't use them for their reliability but because of the cost of conversion. A cluster of servers is more reliable than a mainframe - if you get a hardware problem then you can still continue processing.

Can you imagine trying to cluster a mainframe? Yuck. I know it's done, but the client/server model makes clustering so much more practical. And it costs (relatively speaking) almost nothing.

//JOB1 GOODBYE
//BYENOW EXEC PGM=GOODBYE_WORLD
//SYSIN DD *
//SYSOUT DD SYSOUT=*

thnx, i miss JCL, and COBOL.
i'm shedding some tears right now

Vaneshi: I'd also wonder what they're replacing it with, if the 'rack upon rack of Dells' I'm sure this modern thinking (i.e. could find neither arse nor elbow with a map, torch and labels stabled to each) CIO has undoubtedly plumped for is actually cheaper than the $30kp/a power bill when looked at system capacity. Yes POWER is quite energy hungry, so are legions of Xeon's.

Absolutely! The CIO of NASA could learn a thing or two from you!

/ eyeroll

Ed Grubermann: [2.bp.blogspot.com image 580x580]

Oh great, there goes the mainframe!

Tom_Slick: So much for my AS/400 knowledge getting me a job at NASA.

AS/400 is now System i (System p Hardware now really) NASA just got rid of their last System z. They probably still have some AS/400 systems around somewhere. Despite what the article and many posters here said you can't just get the same performance and reliability of a cluster of Windows or Linux boxes, but I guess with NASA no longer in the business of sending humans into space, they can make some allowances in reliability in the name of saving costs.

\ A bit more progress like that and NASA will be reduced to doing their IT 'in the cloud'

T-Boy: I like that the article tried to explain what a mainframe is, or was.

Agreed. I always have fun trying to describe how computers range from thin clients to desktops to workstations to servers to mainframes, including all of the overlap in terminology.

I always considered a mainframe to be a high end server with a high speed bus that was able to interconnect a massive number of parallel processors and huge gobs of memory, all with insane levels of error detection and correction, the ability to hotswap any component, and software that was highly virtualized. I always differentiated mainframes from big iron in that the former had more of everything - more processors, more memory, more levels of error resiliency.

I'd like to know a bit more about the ability to run legacy mainframe partitions on distributed blade platforms. Are there emulators for running older VAX or Alpha VMS LPARs on commodity blades? Same regarding S/370, S/390 and zSeries LPARs. I always wince when I hear about smaller companies purchasing hand-me-down mainframes because their entire company is based around some legacy VMS apps. You'd think it would be a good thing if they could just emulate that crap on some Itanium system.

ZAZ: Dinjiin

HP has rack mount VMS systems and you might be able to get one new for under $20,000. I think VAX compatibility is part of base VMS on Itanium. I have not confirmed.

Itanium is still on the market in new systems? I thought they gave up on that, or was that just for desktop systems?

ZAZ: Reliable systems take a lot of software effort. I used to work on cluster software (c. 1999) and the test/debug time was much, much longer than planners had anticipated. There is a general habit in software development to skip error testing, but error handling is where you squeeze out a few more 9s of reliability. (We didn't have any 9s before I left that job.)

Thing is, a lot of clustering is out of the box now.

I do Windows/asp.net and adding an extra cluster is very easy. It's basically built into the .net framework.

I've also seen things with Hadoop clustering and Microsoft's Azure in the past few days.

Barakku: Can you imagine trying to cluster a mainframe? Yuck. I know it's done, but the client/server model makes clustering so much more practical. And it costs (relatively speaking) almost nothing.

Horizontal scaling makes a lot more sense because it's all off-the-shelf. I know a company building a massive Hadoop cluster and the first stage, they're using a load of old laptops. If the project takes off, they'll add some rack mounted servers.

I'm sure the DoD still has lots of mainframes. Various defense programs have very long lifespans (from initial design to last use) and if a mainframe was used initially then one will probably be in use to the end because the cost of porting the software and absolutely certifying the accuracy of the port would be too great.

I'm just guessing here but consider the DSP satellite program. The first launch was in 1970 and I think they are still operational today. 25 years ago I was at a ground station that did sensor-processing for one and it was running real-time on a huge old IBM mainframe which I recall was badly obsolete technology at that time.

Often such programs have the mind-set "if it is not broken then don't fix it" and "don't port software, instead hold on until the whole system is re-engineered" .

I wonder how many old IBM360s and 370s the DoD still has under maintenance contracts?

Barakku: Itanium is still on the market in new systems? I thought they gave up on that, or was that just for desktop systems?

Yeah, Itanium is still around. Intel still saves their best error detection and correction for the Itanium, but I hear that the latest Xeon chips may start to get it as well. For finance and medical institutions, the extra cost of the hardware is less than the cost of a lawsuit or negative publicity from a computer glitch.


ZAZ: I think VAX compatibility is part of base VMS on Itanium.

Yeah, but does OpenVMS under Itanium include binary compatibility with Alpha or VAX binaries?

Vaneshi: I'm not saying the article is wrong, yes a zSeries is a frame element no doubt. I'm not totally convinced the picture shown is of a 12 year old chassis though.

I'd also wonder what they're replacing it with, if the 'rack upon rack of Dells' I'm sure this modern thinking (i.e. could find neither arse nor elbow with a map, torch and labels stabled to each) CIO has undoubtedly plumped for is actually cheaper than the $30kp/a power bill when looked at system capacity. Yes POWER is quite energy hungry, so are legions of Xeon's.

You'd think that setting up a solar array in Alabama/Florida/Texas would yield enough electricity to power one of those things. At the amount of money they pay for electricity per year, it could probably pay itself off sooner rather than later.

Tr0mBoNe: Clusters do the job of Mainframes but use Linux or Windows and normal engineers and administrators know how to use them. A couple racks full of blade centres, a large storage array (with a well thought out backup policy) and a big network will do the job. IBM sells all of that stuff. For several millions you can get the performance of a mainframe that cost several hundreds of millions.

You won't get past the first level of interviews for an IT job if you told me that you can just replace our mainframe with a "cluster". Building a bank of systems based on *nix or Windows comes with its own headaches. Maintainability being one of them. And its not always about performance. Reliability is as important if not more so for many applications. And most Windows or Linux sys admins wouldn't know the first thing about building a reliable architecture. Let alone a .net developer. Most don't have the experience or the discipline.

And the cost of hardware is just one component. There is software costs, customization, maintainability, power, DR, etc that all factor into the TCO of an application. For many applications - the mainframe is still very competitive if not cheaper than a "cluster".

/built a linux cluster back in the late 90s when it was still cool

bravian: You won't get past the first level of interviews for an IT job if you told me that you can just replace our mainframe with a "cluster". Building a bank of systems based on *nix or Windows comes with its own headaches. Maintainability being one of them. And its not always about performance. Reliability is as important if not more so for many applications. And most Windows or Linux sys admins wouldn't know the first thing about building a reliable architecture. Let alone a .net developer. Most don't have the experience or the discipline.

You're obviously smarter than those guys at Amazon, eBay and Google. Well done.

So, NASA is telling everybody that THEY are the ones who shut down the mainframe, huh?



/or post a Guy Fawkes mask... whatever floats your boat

Marine1: Vaneshi: I'm not saying the article is wrong, yes a zSeries is a frame element no doubt. I'm not totally convinced the picture shown is of a 12 year old chassis though.

I'd also wonder what they're replacing it with, if the 'rack upon rack of Dells' I'm sure this modern thinking (i.e. could find neither arse nor elbow with a map, torch and labels stabled to each) CIO has undoubtedly plumped for is actually cheaper than the $30kp/a power bill when looked at system capacity. Yes POWER is quite energy hungry, so are legions of Xeon's.

You'd think that setting up a solar array in Alabama/Florida/Texas would yield enough electricity to power one of those things. At the amount of money they pay for electricity per year, it could probably pay itself off sooner rather than later.

That's exactly why Google and Microsoft both have huge data centers near large hydroelectric plants in Washington and Orgeon. All the power they will need and at a relatively stable cost compared to fossil fuels.

Will it dream?

Wait, what about those supercomputers? (Like NASA's Columbia).
Those are mainframes. (well, more like an array of computers, but I digress)

Did it sing "Daisy, Daisy" as it was shut down?

As a former IBMer, I'm getting a kick out of this entire thread.

IBM quit caring about hardware right after Lou Gerstner retired and handed over the reins to Sam Palmisano. Now we'll see what happens under Ginny Rometti. Personally, I think IBM has completely lost its way, and I don't think they don't have the first clue that they have.

//Dude, you're getting a Dell

jaytkay:
Absolutely! The CIO of NASA could learn a thing or two from you!

Glad to see you're in full agreement. I was expecting a snarky response from the likes of Fark but there you go.

just_intonation: As a former IBMer, I'm getting a kick out of this entire thread...Personally, I think IBM has completely lost its way, and I don't think they don't have the first clue that they have.

They totally collapsed without your irreplaceable skills*.

Net profit $5.5B in the last quarter of 2011. The stock price has doubled in the last 5 years.

/ *Belt onion

just_intonation: As a former IBMer, I'm getting a kick out of this entire thread.

IBM quit caring about hardware right after Lou Gerstner retired and handed over the reins to Sam Palmisano.

I would, in all honesty agree. During my sting at big blue I can safely say: They didn't care about software (DB2, AIX, WebSphere, etc), didn't care about hardware (POWER, zSeries, shoddy build quality on the laptops, etc.), didn't care about service providing ("What the customer wants the customer doesn't get" to quote a service manager, "Fark em" to quote another), didn't care about actually getting the contracts (they let a local NHS trust just walk right out the door), didn't care about staff (Projects Meteor & Merlin to name two, plus the whole if you don't spend your training budget then it's your Xmas bonus crap).

I can't tell you what IBM's focus is or what they care about. I don't even think they care about money, they let so much of it walk out the door. I am however surprised they haven't gone totally tits-up.com though.

ZAZ: I am surprised to read that NASA has no mainframes. What do they use when they want a reliable computer?

jaytkay: just_intonation: As a former IBMer, I'm getting a kick out of this entire thread...Personally, I think IBM has completely lost its way, and I don't think they don't have the first clue that they have.

They totally collapsed without your irreplaceable skills*.

Net profit $5.5B in the last quarter of 2011. The stock price has doubled in the last 5 years.

/ *Belt onion

Considering I left in the last few months, I was part of that.

Also note that the vast majority of that net profit is from software and services, not new hardware sales. There aren't many new zSeries being installed. Power is increasing sales in a dying market (and IBM thinks that's a HUGE deal for some silly reason) and xSeries is getting its ass kicked on price by Dell and its competitive edge's ass kicked by HP, and is about to become irrelevant unless something 'major' happens.

So, unless IBM thinks it can survive on software and services alone...

last place I worked at that had a mainframe, we replaced it with a super-mini. The hardware cost 1/10th the annual maintenance fee for the mainframe. Then we had to put a new A/C system in because the old one had so much excess capacity that the temperature was cycling between 20C and freezing.

What unplugging a mainframe may look like.