So, You Want To Buy A Mainframe

The computers we are used to working with are more likely to be at the smaller end of the computational spectrum. Sometimes they are very small indeed, such as tiny microcontrollers with only a few GPIOs. Others are single board machines such as a Raspberry Pi or an Arduino, and often a desktop or laptop PC. Of course, while these can be very capable machines, they don’t cut the mustard in the upper echelons of corporate computing. There the mainframe still rules, sitting in air-conditioned machine rooms and providing some of the glue that cements our economy together.

Most of us will never own a mainframe, even if sometimes we marvel at people who rescue ancient ones for museums. But it’s not impossible to run one yourself even if it isn’t cheap, and  [Christian Svensson] has written a guide for the potential purchaser of a more recent IBM model.

This is a fascinating piece as an uninformed spectator because it reveals something about the marketing of these machines. A fridge-sized rack may contain much more hardware than expected because all machines ship with high specifications installed but not enabled by licensing software. In some IBM machines this software comes on an attached laptop which goes missing when the mainframe is decommissioned, we’re told without this essential component the machine is junk. The practicalities are also considered, such things as whether the appropriate interface modules are present, or how to assess how much RAM has been installed. Powering the beast is less of a problem than you might expect as they ship with PSUs able to take a wide variety of DC or AC sources.

Once upon a time the chance to own one of the earlier DEC VAX minicomputers came the way of your scribe, the passing up on which has ever since been the source of alternate regret and thankfulness at a lucky escape. The ownership of second-hand Big Iron is not for everyone, but it’s nevertheless interesting to learn about it from those who have taken the plunge. There’s a tale unfolding about the ownership of a much older IBM room-sized computer at the moment.

IBM mainframes header image: Agiorgio [CC BY-SA 4.0]

37 thoughts on “So, You Want To Buy A Mainframe

    1. I do not think the Unisys mainframes suffer from the dreaded “missing-SE” disease. They pretty much give away OS2200 and MCP anyway for non commercial “hobbyist” use, so if the console is missing, you could probably rebuild it.

      You do not even need any special hardware anymore, as they have a nice emulator for both the Libra (Burroughs/MCP) and Dorado (Univac/OS2200) lines.

  1. About 6 years ago, I worked for Big Blue, refurbishing mainframes for resale. I didn’t get into the software loads. The linked article brought back memories of cleaning, testing, disassembling, reassembling to order, testing the “new” machine.
    And yes, if you don’t have the “software/firmware piece” that defines the machine, you have nothing.

  2. A note on the HMC’s – you need one of these for most non-Z machines (IBM E or S POWER series – probably others) – they allocate the hardware to a partition. You can’t just run it ‘bare metal’, you still need an HMC to assign all hardware to a single (or multiple) partitions, if that was your intent. — You can run a virtual HMC these days to manage this, but not sure if you can get a hold of an image for that in the wild without licensing. Then again maybe S/E series servers wouldn’t be considered ‘mainframes’, so disregard :-)

      1. Correct.
        If it works, don’t upgrade.
        On a tour a while back, an IBM mainframe sitting there churning bits and bytes, I noticed the on/off power switch was the same as the one used in the very first IBM PC.
        I had to chuckle. LOL

        1. Err, no.
          Maybe the _look_ the same, but rest assured: the internals aren’t.

          That “simple little on/off” switch on the mainframe is way more than “a simple” switch. Sure, the function it delivers is on/off. For what you can observe from the outside. But there is quite more to it implementation wise.

    1. Well, consider the maximum you can put into a contemporary mainframe – terabytes of RAM and hundreds of cores, as well as hundreds of PCIe cards – and the mainframe quickly becomes a force to be reckoned with. Then add on that this are extremely reliable (the memory is redundant for example, and you can hot swap CPUs and PCIe cards as far as I know) as well as you don’t have to pay everything up front but can use cloud-like pay for what you use models.

      Sounds pretty cool and modern to me.

      The article is about a 8 year old mainframe that had the entry level amount of RAM at the time, so it is pretty underpowered by today’s standards.

    2. Keep in mind this isn’t just a Pentium in a bigger box. Sure, bytes are bytes when it comes to storage, but things like CPU frequency don’t necessarily track with the systems you’re used to evaluating.

      These machines are built with redundancy, reliability, and massive parallelism in mind. The parallelism is both in the form of smart job/multithread management, and in terms of partitioning, letting you carve off virtual slices of the machine to run independent workloads.

      But there are also special features built into the bare metal to better support enterprise workloads. For instance, I was told early in my career that on some models of the AS/400 (not quite mainframe, but works for this example), a database row fetch is one instruction. Compare that to MS SQL Server where it clocks in at 1000+ instructions.

    3. Disclosure: I work at IBM.

      Specs are not everything. If you compare only GHz and memory, you are not seeing the big picture, and that matters more than specs.

      I worked on a 700MHz, 16GB mainframe on 2005. My desktop was the top of the line Athlon XP 2600+ that days, and it could read around 3.5GBytes/s on RAM. That 700MHz mainframe did read the disk at 4GBytes per second. It could initiate around 3k SSL VoIP connections per second too. Processor speed is important, but only comparing similar architectures, and a mainframe processor share few similarities with a PC processor.

      For example, on a mainframe any instruction reading RAM is considered IO, a slow instruction, so the processor issues the read, suspends the process and goes to the next. The processor does not waste time reading RAM. There is a dedicated processor for that, and this processor will talk to the other processors (yes, there’s a star network of memory management processors) if anyone have that address on their cache. If any have, they will share the data, and the memory management process will put the data inside the main processor cache, and interrupt it. And the cache sizes are enormous. I know there is a tradeoff between cache size and cache speed, but there are several levels of cache, and that nice memory access processors taking care of that. Kind of a CDN for memory, implemented in hardware.

      IO processing is another thing. There are special processors for that too. And the IO bus is not a bus, is a subsystem, with memory, processors, firmware, queues. IO processors interact with each other. They have priority queues. And the processor is as powerful as the main processor.

      You need crypto? RSA, AES, SHA512, HMAC, hardware generated random numbers? Well, there’s a subsystem for that too. A dedicated cryptographic card doing crypto on hardware.

      So the processor only executes instructions. Does not access memory, does not manage its own cache, does not manage interrupts, does not do any IO operation, does not care about crypto. It only executes instructions, and that makes a lot of difference.

      Redundancy is another level too. You can replace the processor with the mainframe running, without any downtime. Memory banks too. Every single component of a mainframe can be swapped without shutting it down. Every one, from cabling to main processors, to cooling, power source. Every adapter card have 2 bus lines, connected to dual IO switches, dual power lines to dual power supplies, dual batteries. Everything is at least duplicated.

      Today this changed, but back in 2007, even the instruction processing were duplicated. Two processors would execute any instruction, and compare the results. The same? Go ahead. Different? Another processor would execute the differing instruction and compare the results, and the processor with the wrong result were put offline and marked as faulty. IBM discontinued this feature because there were no execution errors in a long, long time.

      It’s expensive, sure, but if you are on a company that every minute of downtime costs hundreds of dollars (a bank, a large retailer, telecom…), the price is offset pretty fast when you spend years without unexpected hardware downtime.

  3. I had a coworker who bought the house next door and filled it with Good Stuff he got off eBay or somewhere (he was obviously unmarried). I helped his sister clean it out when he died. Down cellar he had a HP 2100 in apparently working condition. He’d worked on one at the lab (Air Force Research Lab) for years. I wrote some computer museum and they told me they had literally tons of them but I could ship it there at my expense. It took me hours to take it to bits and dumpster it.

    I’m thinning out my own collection. If I can’t sell it and haven’t used it in ten years…. OUT. Might take a picture though, it’ll last longer.

    1. Shame it went in the trash, but for most of us space is too limited to store large items that have limited practical use. You want to save it all, but you can’t. I remember last year when Dave Jones (EEVBLOG) went to save some stuff from a warehouse a computer museum was renting that was going to be demolished(https://www.youtube.com/watch?v=mK4M8dv4NNU.) Some interesting things in there for sure but they tried to save everything including the kitchen sink. If the warehouse wasn’t going to be demolished most of everything there would likely have sat for decades collecting dust.

  4. My old place of employment had a DEC VAX780 and associated peripherals (TU tape drives, HA disk arrays, CI’s etc). Took most of it home (was single at the time), filled up my huge living room. Then realized, okay now what ? The entire system needed HUGE ass amounts of electricity (3 phase no less), generated tons of heat, lots of noise and I only had 2 LA-36 Decwriter console terminals, and 1 VT220 terminal for user I/O. Also got the DEC diagnostic tapes – again, now what ? Ended up scrapping it all (which is what I employer wanted to do in the first place.. lol)

  5. One thing I learned the Hard Way – always double-check the physical specs. About 16 years I ago I wanted to improve my Solaris skills and bid on an E250. To put in a one bedroom apartment. I really shoulda checked the specs better ahead of time.

    Ok, not really much of a “mainframe” but I think it counts. Barely.

    1. Ah. Those were big guys. We were still using them as our main production in my SAAS shop when I quit in ’07. Just a micro tho – most of the same capabilities packed in to the 4Uish e220 (except for the large scsi backplane and redundant power), and the totally desktop Ultra 60. Two four-hundred-ish MHz processors and 2gb of RAM as I recall

  6. I had a DEC 10000 at home… also not really a main”frame” but large enough to act as table for the terminal laptop. Needed 4 People to get it in here. fondly remember the 5 1/4″ full size, 800MB discs spinning up (i took one apart – 11 platters!)

  7. Since ever I limited myself to hoard “desktop” variants only.
    So I’m proud owner of VAXStation 3100/76 and VAXStation 4000-60 (among others, of curse -pun intended)
    Yes, far away from Mainframes – but still Real Computers [TM] of the kind I learned to “walk and chew” on…

  8. I once owned a PDP-8E in a PDP-8 F box that was hand-constructed by a couple of DEC employees out of the easily accessible DEC scrap pile in Maynard. I didn’t have even a TTY, just the front panel for control. I made it blink the lights in cute ways. Gave it to a fellow (single of course) who had a computer museum full of other DEC junk in his two-bedroom place.

    For every genuine computer historian there are probably five dozen of us madmen who own old computer junk. My wife and I also have old trucks, old horses and old saddles. The old ones are the best. Can’t say that for computers.

  9. Built In Upgrades”? Too soon?
    Ordered an Intel “Small Server” mobo, discovered too late, Max ram 2 gig tops. WTF?
    As I recall, Sony shipped a laptop with 2 gig max ram..
    Apple too, iMacs shipped with 2 gig.
    HP seems to be Ok, my worn out Z800 came home with 48 gig ram, and room for more..
    HP Z800 even shipped with hot swap SATA, AND SaS ports. Built in hardware R.A.I.D.
    /rant over

  10. The closest I ever came, was about 25 years ago, when a buddy bought a full height rack at an equipment auction, and then called me up to ask “Do you want any of the junk that’s in this rack?”. It seems he just wanted the rack. So, I raced over to find that he was giving me a complete PDP-11 system, vintage about 1976. That was the good news. The bad news was that it was a PDP-11/03 model, which had 4K of SRAM installed, and was the industrial control computer, without the fancy lights and switches. It also had a pair of RK05 disk drives (14 inch), with one being the standard removable disk, while the other was the fixed disk variant, which held 2.5MB and 5.0MB respectively. Oh, and did I mention that they were HEAVY? I kept it for a year or two, mostly as a curiosity, before I gave it to a PDP-11 collector, who set a new land speed record when he heard that I was giving it to him.

  11. Would it be possible to acquire one of these to use for gaming – as in running multiple instances of Nox of a mobile game? I can run 5 on a windows laptop – but with a mainframe……

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.