Migrate From Mainframe? To What?

From Joe Clabby, www.clabbyanalytics.com

Gartner, the industry’s preeminent information technology (IT) research and analysis firm, has published several reports and case studies over the past few years that promote the idea that IT buyers should migrate their applications off of mainframes and move them to other, more “modern platforms”.  Part of Gartner’s logic, it appears, is that there is an impending-doom shortage of mainframe managers that is about to occur as elderly mainframe managers retire — so Gartner implies that moving applications to other “more modern” platforms might ensure the long term viability of enterprise applications on those platforms.

I have two major issues with Gartner’s perspective and its recommendation:

  1. Where is the proof that mainframe skills will decline to critical levels over the next several years?  And,
  2. Which “modern platform” is Gartner advocating?

On the Alleged, Pending Skill Set Crisis

My perspective on Gartner’s suggestion that IT executives consider moving their applications off of a mainframe due to an alleged, forthcoming decline of mainframe skills promotes an “urban legend” (something that makes sense on the surface but has no basis in fact).  So, in a counter opinion, I demonstrated that mainframe skills can be found in abundance in various geographies — and I showed that there has been a huge increase in the number of students studying mainframe management, and in colleges and universities teaching mainframe curriculum (in 2005, 10,000 students were taking mainframe courses — that number is now 50,000; and the number of institutions teaching mainframe courses has tripled over the same time frame).  Further, I provided analysis that showed that the mainframe resource pool may number 400,000 people.  I then asked Gartner to prove that there is indeed a professional skills crisis in the mainframe world.  I’m looking forward to Gartner’s response.

On Gartner’s Migrate Advice to Migrate to a “More Modern” Platform

As for Gartner’s advice to move to a more modern platform, I challenge Gartner to:

  1. Show me a more modern commercial platform than a mainframe; and,
  2. Clarify which “more modern platform” Gartner has in mind.

With respect to Gartner’s contention that there are other, more modern platforms than a mainframe, I observe that even Gartner admits that mainframes run modern JAVA workloads, that mainframes are good for large Linux consolidations, and that mainframes are strong in SOA (service-oriented architecture) message handling.  And Gartner also points out that mainframe specialty processors (zIIP and zAAP) make mainframes attractive for handling certain data-intensive and JAVA workloads.

To Gartner’s own list of advanced mainframe characteristics allow me to add the following:

  • Name any other system in the marketplace that has a more advanced virtualization, provisioning, and workload management environment than a mainframe — or that can operate at 100% utilization for extended periods of time;
  • Name any other system in the market that has more advanced security and crypto services than a mainframe (note: no other “more modern platform” offers EAL level 5 security);
  • Name any other commercial system in the market that has a larger communications backplane than a mainframe (this is important for handling massive amounts of data and for process millions of SOA messages — and it cuts down on the need for hundreds of network interface cards and associated bridges, hubs and routers);
  • Name any other system in the market that provides better meantime-between-failure and greater business resiliency than a mainframe; and,
  • Find me a system with better linear scalability than a mainframe.

In short, I believe that mainframes are the most modern platforms available in the commercial marketplace today.

Still, several of Gartner’s research reports seem to indicate that there other, more modern platforms that can serve as alternatives to mainframes (although Gartner doesn’t name them).  But maybe some of Gartner’s marketing activities provide a clue as to which platforms are their “more modern” mainframe alternative choices.

  • It did not escape my attention that Gartner has participated in several Hewlett-Packard (HP) mainframe migration road shows over the past few years.  At these road shows, HP has promoted the idea that Itanium-based servers can be used to replace mainframes.  I see Gartner’s participation in these promotion tours as an indication that one of the “more modern” platforms would be an HP Itanium-based server.
  • I note that Gartner recently participated with Microsoft in a mainframe migration program in Denmark.  Because Windows only runs on x86 servers (Microsoft has just withdrawn future development on Itanium), I assume that Gartner is also promoting the idea that migrating mainframe applications to Windows servers is a good idea.
  • I found no proof that Gartner has promoted the use of either Oracle/Sun UltraSPARC servers or IBM POWER-based servers as mainframe replacements.

Given these observations, I think past Gartner activities indicate that the company believes that Itanium servers present a more modern platform alternative to mainframes.  If this is the case, I note the following:

  • From my perspective, HP’s Business Critical Systems Division (BCS) is in big trouble (this is the division that sells HP’s Itanium-based servers).  Over the past year, this division has recorded quarterly drops of -22%, -33%, -30%, -29%.  And given that HP sells about 85% of all Itanium servers, these sales figures show us that Itanium is in big trouble.
  • As of version 6, Red Hat has withdrawn its RHEL (Linux) on Itanium.
  • Microsoft has announced that it will no longer build Windows for Itanium.
  • Fujitsu (a prominent member of the Itanium Solutions Alliance) has announced that its future products will be Xeon-based (for more details, see this report here).
  • VMware (the x86 market leader in virtualization software) long ago decided not to offer its virtualization stack on Itanium.

To me, the Itanium microprocessor is in big trouble due to the overlap of Intel’s own Xeon with Itanium architecture.  Further, because HP’s Itanium sales have experienced such large drop-offs (HP sells about 85% of all Itaniums), I’d argue that Itanium sales as a whole are in big trouble.  And, because Microsoft and Red Hat have ceased Itanium development —and because I see the potential for other vendors to pull Itanium support — I’d argue that the Itanium ecosystem is in big trouble.  With all of these troubles, I don’t think I’d be thanking Gartner if the company had recommended that I move from my stable mainframe environment to an Itanium platform.

In a effort to be a little balanced when it comes to Itanium architecture, I’d like to point readers to this report by Timothy Pricket Morgan where he postulates that “there will be Itanium processors available for at least six to seven years” and he explains why.  I think Mr. Prickett Morgan’s report is one of the most balanced pieces I’ve read on Itanium over the past several years and it is worth reading.

Gartner’s recent involvement in a Microsoft application migration road show in Scandinavia also indicates to me that they may now believe that Intel’s new x86 multi-core Xeon’s represent a viable, more modern alternative to a mainframe.  And, from my perspective, x86 multi-cores are excellent processors that offer a well-balance architecture for computing serial, data-intensive, and parallel workloads.  But, I have yet to see a PC architecture successfully replace a mainframe (consider what needs to be done to a PC architecture to make it as secure, reliable, and resilient as a mainframe — and this may be why PC servers don’t rival mainframes today).  So, until I see an x86 server environment that can rival a mainframe in security, reliability, and resiliency — as well as handle general workloads in an optimized fashion — I’m going to continue to argue that mainframes are the most modern architecture for enterprise computing available today.

I have to admit that I found it curious that I could find no Gartner documents that promoted the use of IBM’s POWER servers as “more modern platforms” — especially with all of the migration activity that is taking place from Sun/Oracle and HP to POWER platforms.  In IBM’s first quarter, IBM’s Migration Factory indicated that a total of 107 customers selected IBM Power Systems over Oracle/Sun systems, and 41 chose IBM over HP in the first quarter.  (In addition, IBM’s x86 multi-core System x recorded 21 migration wins — while IBM’s System z recorded 16 wins — all against either HP or Oracle/Sun).  This statistic is pretty interesting because it shows IT buyers moving away from “more modern” platforms to mainframes.

What Clabby Analytics Would Recommend

Instead of taking Gartner’s advice and migrating your applications from mainframes to other platforms due to a pending (but unsubstantiated) shortage of mainframe skills, you might want to do an inventory of your applications.  Look closely at the threading characteristics of your applications (the number of independent processes and their memory requirements).  And also examine the impact of constantly driving messages and threads over a network (as opposed to driving them over a large, high-speed internal bus).

As you consider application migration, consider these points:

  • If an application is parsed into many distinct threads that can be processed on separate servers and the results can then be easily reassembled to produce a final outcome – and if each of those threads does not require greater than four gigabytes of memory to execute – then that application is a candidate for a scale-out computing design (such as x86 blades, racks, or towers).
  • If an application consists of numerous, tightly-coupled threads (threaded with a threading library) and can benefit from sharing large amounts of memory, then that application should be run in a scale-up design (such as high-end Unix/Linux server environments; new, memory-rich, scale-up x86 multi-cores; or a mainframe).  Memory use is important because the more data that can be place in main memory, the faster it will get processed.
  • From a network design perspective, consider the impact of deploying your applications in a distributed computing design versus a self-contained, scale-up design.  If you choose a distributed computing design, be prepared for increased networking costs (for hubs, bridges, routers, NICs), and increased systems/network management costs (because it takes more people to manage distributed systems environments than scale-up environments.
  • Also from a network perspective consider this: computing infrastructure is standardizing on service-oriented architecture (Web services standards-based program-to-program communications).  Service oriented architecture is a message-passing architecture that generates a lot of network traffic.  Clogging you network with all of this traffic creates latency issues.  Scale-up designs, however, have very high speed internal buses that don’t get clogged — even with high volumes of message passing.  Most Sun users have heavily deployed distributed computing environments.  Now, during a migration, it may be time to rethink whether distributed computing is really the best design point for your data center of the future.

As you weigh your application design points, also weigh reliability, availability, and security (RAS) requirements for your applications.  From a RAS perspective, both scale-up systems and scale-out blades are known to be reliable.  And should a failure occur, both systems have the ability to fail over either to other servers or to virtual workspaces within the same scale-up platform or blade enclosure.  In the area of security, there are distinct advantages to scale-up platforms (in scale-up environments there are fewer access points that need to be guarded ─ and also, less data is sent over a network where it has the potential to be intercepted).

Also consider manageability if you chose to migrate from a mainframe to another architecture.  From a manageability perspective, it can be argued that having fewer servers to manage is easier than having many servers to manage.  And as proof, consider that large, scale-up mainframes require significantly fewer managers than scale-out environments consisting of hundreds of servers.

Summary Observations

For those who read Clabby Analytics research reports on a regular basis, you know that we believe that the midrange and high-end server markets are consolidating around three architectures: x86 multi-cores, IBM POWER, and IBM z.  We have reached this conclusion for a number of reasons, including:

  • The ascendancy of Xeon multi-cores into the midrange and high-end;
  • The impending collapse of the Itanium ecosystem;
  • Migration trends in the Sun UltraSPARC platform base;
  • The strength of the new version of IBM’s POWER architecture (POWER7); and,
  • The strength of IBM’s System z mainframe environment (mainframes operate several of the world’s largest economies and drive several of the world’s financial and retail systems — mainframes are going to be with us for a long, long time).

And for readers who are familiar with my reports, you know that I strongly advocate the use of mainframes in situations where high-availability, business resiliency, and security are of paramount importance.  Further, I recommend mainframes for handling data-intensive workloads as well as for performing massive amounts of transaction processing.

I do acknowledge, however, that there are situations where moving applications off of a mainframe is appropriate:

  • Your workload has decreased and you can’t keep the mainframe fully burdened (mainframes are workhorses – and to maximize ROI you need to keep them running at greater than 70% of capacity).  If you can’t keep your mainframe busy, another alternative (like an HP Integrity server, or an IBM Power System, or a Sun UltraSPARC system) should possibly be considered.
  • You are running an old mainframe; haven’t updated your hardware for years; and are paying increased maintenance costs (“more modern” servers look very attractive under these circumstances [but remember to compare a more modern mainframe to these servers]);
  • Your company has been acquired and wants a uniform information systems infrastructure;
  • You have found suite of applications that your company likes – and that suite doesn’t run on mainframe architecture; and/or,
  • IT management has a distributed systems bias (i.e. ─ they like to deploy, manage and tune distinct application servers, database servers, mail servers and so on.  Some IT buyers like to design their Unix/Linux-based around dedicated servers and then focus on raw performance).

I do not, however, agree in any way, shape or form with the contention that moving applications off of a mainframe due to an alleged, impending skills shortage is a good idea.

Nor do I believe that migrating to certain “more modern platforms” without close scrutiny of an application’s design and RAS requirements is a good idea.  I would like to see Gartner withdraw its advice to move off of mainframes due to an alleged, impending skills shortage or substantiate that skills shortage will indeed occur.  And as for “more modern” platforms, I’d like to know specifically which “modern platforms” Gartner is advising its clients to move to (and I sure hope they’re not suggesting Itanium and UltraSPARC)…

Copyright Clabby Analytics 2010. Used by permission.

Clabby Analytics is an information technology (IT) research and analysis firm that produces research reports and opinions that we base on customer case studies and our competitive analysis. Unlike a lot of research and analysis firms, however, we tend to overtly take sides and advocate for positions that we believe in. We don’t pretend to be unbiased — in fact we have very clear biases toward centralized mainframe architecture, scale-up Unix/Linux architectures, and blade servers. Our research reports advocate these positions, and we encourage our readers to consider what we have to say — read what other research analysts have to say — and then arrive at their own opinions and conclusions.

About Merv Adrian
Gartner Research VP, technology analyst and consultant, 30 years of industry experience, covering software mostly, hardware sometimes.

20 Responses to Migrate From Mainframe? To What?

  1. jragsdale says:

    Great post! People have been ‘hating’ on mainframes since the 80s, yet they still are the power behind financial markets and other massive data applications–doing things no one else can. My personal issue with mainframe apps were the atrocious UIs, but luckily green screens are a thing of the past, and now mainframe apps have Java wrappers (or whatever the buzz is this week) to create highly usable applications.

    I’ve talked to lots of companies who are doing more enterprise knowledge management initiatives to capture the learnings of older workers so ‘brain drain’ won’t impact customers as people retire. We’ve been hearing about this for a decade now, surely everyone has a strategy in place. If not, shame on you.

  2. Merv Adrian says:

    Hey, John – thanks for the comments. I loved Joe’s pieces so I asked and he was gracious enough to let me post them here.

  3. Great blog. Sober, but with all the right arguments. I am missing 1 important point I think we should make.
    Why is the mainframe still seen as expensive by a lot of people?
    And more important; why is the Mainframe cost centre often charged with things that do not belong there?
    Examples: I have seen Energy charges of > 50% go to the Mainframe. In 1 occasion because “the mainframe is running 50+% of all workload”… Duhh.. In another occasion the Mainframe DID use 70% of the datacenter energy 20 years ago and we just left it like that….. Even though we have 12,000 distributed servers….
    I have also seen charges for Project management Costs to the mainframe cost center of > 50%. While there was NO new project running for the mainframe.
    Is this because we want to make the Distributed Environment look cheap by making the Mainframe look expensive????

    I would be interested to see your comments

  4. If there’s a shortage of skilled people, you can always overcome that with appropriate training. Ahem.

  5. Merv Adrian says:

    A bit self-serving, Steve, but thanks.

  6. Merv Adrian says:

    I’ll leave it to Joe to get into details, but over the years alternate platforms were often acquired at smaller individual prices and took on only a part of a company’s workload, app by app. Consolidation has changed the dynamic, but thinking has not necessarily reflected the economic shifts implied.

  7. Terry Draper says:

    One aspect that is overlooked is the excellant performance and problem resolution tools on the mainframe. Nothing gets anywhere near monitors such as RMF (and its mainframe competitors).
    From what I have observed the performance resolution for other platforams is add another (or many) processors or even boxes.
    With nearly 40 years in mainframe I cannot see any of the ‘modern platforms’ getting close to the mainframe when all the true qualities are considered.

  8. joeclabby884 says:

    First, thanks for all of the comments. This article was meant to start a dialog — and it did. On the cost question, its pretty simple — mainframes do cost more from an acquisition point of view. But start rolling in all of the additional monies for management, security, licensing, resiliency, adaptors/connectors, real estate (for sprawl), energy consumption, and so on — and mainframes may actually cost significantly less than the so called “industry standard” server solutions. But don’t take my word for it. Here are two URLs to mainframe customers who say paying the extra acquisition cost money is worth it.

    Click to access CornerBankFinal.pdf

    Click to access Banco_de_Credito_final.pdf

    On the question of knowledge capture — check out CA Technologies’ new CA Mainframe Chorus. Wow! It simplifies mainframe management and captures knowlege for future generations of mainframe users. This is the kind of software that IBM mainframe shops should be using NOW.


  9. JimboP says:

    Another reason that there is (or will be) a shortage of mainframe managers is the ITIL-based emphasis on “a manager is a manager is a manager”. This is the thought process behind my new position, starting today, as a Unix manager after 24 years as a mainframe systems programmer (9 as manager).
    With the advent of the “zNext”, I don’t see any other platform as more modern and I think I will be sorely missing some great opportunities to implement new features that no other platform even has.
    All because “a manager is a manager is a manager”.

  10. This is a very nice deconstruction of the conventional thinking that Gartner has been espousing for a number of years now.

    Whenever someone starts mainframe bashing I usually steer the conversation toward a discussion of the requirements of the computing projects and infrastructure that is needed to support the business needs. A nice offhand remark that usually brings a smile is “if you plan to plough a field and you have a choice between an ox (mainframe) and 32 chickens (intel servers) which would you choose?”

    Re: cost, what other platform offers variable workload pricing like a mainframe?

  11. Merv Adrian says:

    Craig, welcome to the blog – great to have you here! I’m a fan of your work, too – the quarterly database review you just put up at http://tdan.com/view-featured-columns/14063 is a nice piece of work!

  12. Ray Saunders says:

    Interesting that after decades of mainframes, NYSE moved off them. Originally to Tandems before HP bought them out – don’t know what they’re using today, but if they’re still on Itanium, they maybe be wishing they’d stayed with IBM.

  13. Lee Sargent says:

    I have found it interesting over the last several years of moving to “more modern architecture” that every time there is a problem the initial problem is blamed on the mainframe. Almost every time it has turned out to be something else. The one advantage in the blame game is that it is relatively easy to determine the truth or not in minutes because of the excellent tools readily available for diagnostics. (The reverse has not been true-sometimes it has taken days to determine the problem source off the mainframe. Some times it has taken a long time to determine which area has the problem after the denial on the mainframe.)

    The only problem is that people seem to remember the mainframe blame even when it is proved false over and over again.

  14. Merv Adrian says:

    I haven’t seen this myself, Lee – perhaps it’s something that has occurred in the shops you’ve been in?

  15. Mike Chung says:

    Thank you, Merv.

    This article provides good ammunition to counter the madness of so many unnecessary migrations and transformations from reliable mainframes to ‘modern’ platforms. As you’ve pointed out, I have yet to witness a platform that can match the reliability, performance and close to 100% availability of mainframes. But the market, at least here in Europe, is far from rational in this perspective. I’ve seen migrations from perfectly operational mainframes to non-performing platforms based on this so-called Service Oriented Architecture for no other reason than a vague perception of shortage of skilled administrators. Of course resulting in expensive ever-lasting projects, increased IT complexity and an angry business.


    Mike Chung

  16. Al T. says:

    1 comment:
    – The number of individuals with mainframe skills is less important than the number of current and project job openings for those skills. Regardless of how many mainframers there are (and there are plenty) they will disappear if they can’t be absorbed into the market.

    1 Questions.
    – Does any one know of a mainframe installation that involves 2 or more geographically dispersed data centers (say at 300 miles plus) where transaction processing failover occurs automatically if one site is inaccessible and the other site picks up the traffic? If one exists, what is the user experience? How long is the outage if any outage occurs?

  17. bouffon says:

    You listen to talk, and you revel in a technology that is outdated. You speak of stability, reliability and power while legitimizing the energetic balance and especially the monstrous financial cost. Obviously, it is now possible to build simple solutions based on X86, stable, reliable and powerful, and that the financial cost is much less important. The cost of a mainframe solution with its software is counted in millions of euros per year, while the solution X86 runs into hundreds of thousands of euros per year. Until now, large companies such as IBM, have frightened business leaders explaining that only the mainframe could solve and secure their needs. In fact, the skills around java and redhat are there to a point where it is possible to have the redhat on the mainframe. I am very intelligent, no technology resists me, and I have already and I still get out the mainframe replacing it with X86. I have over 300 clusters under my belt, with distances of several kilometers, with parallel databases, active / active, and no problem to manage and splitbrain splitSAN. When I re-read your article, I laugh, in force to hear themselves talk we end up believing their own lies.

  18. Merv Adrian says:

    Well written, I must say. Some things still do belong there, though, and others will take time and skills to move. Inertia is a very powerful force.

  19. Joe Clabby says:

    To bouffon’s comments, I must admit that I’m a bit dumbfounded. Mainframes are outdated technology? Have you looked at which system offers the fastest processor in the world with the most onboard memory cache? Do you know that mainframes are the only commercial servers to have attained EAL Level 5+ technology? Are you aware that mainframe virtualization is a decade ahead of x86 in terms of features/functions and managability? And, as for cost, are you aware that it can cost almost $1.5 million less to run the same banking application on a mainframe as compared with an x86 environment (see my report here for further details: http://www.clabbyanalytics.com/uploads/z_VirtualizationFINALRev.pdf). Also, when your reading it, consider that I did not include how much more it costs to power an x86 server farm, nor did I include the hundreds of thousand of additional dollars it would cost to manage that farm. Nor did I account for all of the additional cable (and associated labor to deploy the physical plant).

    Please go back and read Craig Mullin’s comment above. In the end it is all about which workloads execute best on which servers. As Craig puts it: “if you plan to plough a field and you have a choice between an ox (mainframe) and 32 chickens (intel servers) which would you choose?”

    Finally, I discuss the differences between mainframes, RISC, and x86 environments in great detail on my new site: www:workloadoptimization.com. What you will learn on this site is that some workloads belong on x86, some on RISC, and some on mainframes because each server has characteristics that support workloads differently. Further, there are very significant cost implications to choosing the wrong server. To your original point: x86 architecture does not do every job most optimally — so to argue that a mainframe is old and expensive, and that x86 is au currant and exciting, the way of the future, and the only choice the industry needs — is just plain tiresome…

  20. Jazz says:

    But the points have been made that distributed computing far outweigh the so called centralized mainframe idea- and that most business – or govt agencies are often distributed in nature. Distributed transactions ,however, are not without flaws.YET- seem so far to be doing a better job. Think of all the spaghetti code running on that mainframe, too- If someone rewrote it into something more sane- costs could drop…from what Im reading on the z–this isnt possible. Its assembler or cobol or similar ONLY. If you cant use more modern structured programming- even java- how do you fix the spaghetti code? You cant. Workhorse or not- waste is waste.Of course this is just an overview…what New Jersey is asking for sounds like a “record structure update”.Focus is on C and VC and .NET/ASP and CSharp, etc…not COBOL like it used to be.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

<span>%d</span> bloggers like this: