When you are setting up your system landscape for large and/or multiple application deployments, do you consider mainframe? If not, why not? If so, what factors are you considering.

If you take a real TCO look at large ERP and/or consolidated application landscapes, mainframe is actually quite cost-effective.

My own consultations have included recommendations for scale-up/mainframe/mid-size systems on some specific needs. Honestly, I've never had a customer take said recommendation, rather defaulting to countless scale-out VMs on Intel boxen (in non-trivial cost) and yet still to this day have system management and performance issues.

Curious your take on this. We need to remember that the virtual machines we manage (and apparently love in IT departments) today have been done for decades on mainframe. Most mid-size and mainframe shops have small fractions of support persons managing larger and more complex applications.

Your thoughts appreciated.


It seems to me that you're doing more to express your opinion ("If you take a real TCO look at large ERP and/or consolidated application landscapes, mainframe is actually quite cost-effective.") than really ask a question. On that basis, I'm tempted to vote to close, but won't.

As for the question (to the extent there is one), I'm going to assume that by "mainframe", you mean something like an IBM z series machine, rather than (for example) one of the big Fujitsu SPARC boxes.

I think for many people, it's hard to recommend mainframes for a couple of reasons.

  1. Every vendor has TCO numbers to "prove" their product is the best. Why should somebody trust numbers from IBM more than from Microsoft, Oracle, etc?
  2. Even if a mainframe really would make sense, such a recommendation is unlikely to be taken seriously, and whoever made such a recommendation will often be treated as a pariah.
  3. Even if the TCO would theoretically work out better in some respects, introducing machines with which existing staff are thoroughly unfamiliar would often be a mistake anyway.
  4. Using a mainframe would often lose versatility in other ways. Just for example, an Intel box can easily run Windows Terminal Server to make Word and Excel available, which is a bit tougher to do with a mainframe.
  • Fair enough on the TCO comment, I only put that in there, because I expected that most people would immediately reply with "because it costs so much!"
    – Jé Queue
    Sep 20 '10 at 4:53
  • The TCO #s were actually my numbers, not IBMs. "periah?" Is that what you you would think? Massive virtualization could play quite well with social sites. As for existing staff, the same argument can (and is) played out with introducing new programming languages and techniques to staff programmers. Change happens.
    – Jé Queue
    Sep 20 '10 at 4:58
  • @Xepoch: "Pariah" might be overstating it a bit, but you get the idea -- even when/where/if it would make sense, it'll be rejected out of hand and considered a negative reflection on whoever suggested it. Yes, change happens, but you often need quite a large advantage to justify significant change. Sep 20 '10 at 5:04

If the client already owns and operates a mainframe then deploying a new application on said mainframe often makes sense.

However if the client does not already own a mainframe than you would need a really compelling case to overcome the disadvantages:

  • New operations staff and procedures.
  • Specialist sysadmin staff ("Systems Programmers" - even the job titles are different).
  • Specialist programming staff -- the awful JCL and even worse VTAM are still undead!
  • A massive software bill before you can run your first app.
  • An even bigger software bill for the third party software to run your apps efficiently.
  • 1
    Useful addition. The cost of a mainframe is not (just) the sticker price, but the ecosystem.
    – MSalters
    Aug 10 '12 at 7:51

Not the answer you're looking for? Browse other questions tagged or ask your own question.