The Real Story about the IBM Mainframe Makeover

Since 1999/2000, IBM mainframe has gone through multiple hardware changes each with a different architecture.  HP Superdome has been shipping in 2000, and through in-chassis upgrades, delivers industry leading performance today.¹  

Over the same time period the IBM platform has gone through 3 platform iterations:² 

1. S390 (G5/G6)
2. zSeries (z800,z900,z890 &z990)
3. System z9 (z9BC, z9EC)

On top of that, IBM will announce another mainframe in late February 2008. 

The Superdome has continued to occupy the high-end of the server market with a broad base of industry standard benchmarks, unlike the IBM Mainframe, which has not performed an industry benchmark for years.³

It sounds good, but are all applications on the 3900 servers moving to the mainframe?  A recent analyst report states:⁴ 

“An interesting aspect of this server consolidation is that today only about a third of IBM's applications are running on System z, and those only represent 1% of their total physical servers. From the remaining two thirds, much of the software runs on platforms such as DB2, Domino, WebSphere, or Web servers. While it would be a bit flippant to suggest that these sorts of applications can "just move" from one platform to another-minor tweaks will often be needed to handle version differences or even just changes in physical location-neither will they tend to have particularly deep ties to a given operating system or server architecture. In all, IBM looked at about 16,000 server images when planning this project. They determined that it didn't make sense to move about half; the other half were "possible." Ultimately at least some of these will end up on updated System x or System p servers rather than mainframes. (Although the early phases are predominantly focused on Linux and z/VM, z/OS and applications such as SAP are also within the scope of the project.)"
Source:  Illuminata Report, “Big Green and Big Iron”, October 2007

Meanwhile, HP’s IT Transformation is expected to lower the IT budget from 4% of revenue to 2% while delivering more to the business, with more up-to-date information for faster decision making, and lowering risk to the enterprise. ⁵ 

Reducing IT spend is a continual theme within IT, doing more with less is a constant goal.  System acquisition costs must be justified.  A Mainframe Processor can be used as either a General Purpose processor or a Specialty Engine (IFL, zAAP, or zIIP), but it is still a processor.   The Specialty Engine (processor) on a z9 EC is $125,000 versus $31,500 for Superdome 2 core Intel Itanium 9150 processor.  The HP Integrity Superdome processor is about one-quarter the price of the IBM Mainframe processor.

Why then is the Specialty Engine justified as a cost effective?  Since a Specialty Engine is 9%⁶   the cost of a General Purpose(GP) processor, that means a mainframe GP is $1.3M, closer to $1.4M … for a single mainframe processor!  44 times the cost of a Superdome processor.

Mainframe Memory is not much better.   Since 2005, the cost of mainframe memory has been $8,000 per GB.⁷

A growing shortage of skills and talent in IT is an increasing worldwide threat.  Companies may be surprised to see the retirement of their experienced mainframe staff cause an issue in their data center in the next several years. Though IBM’s zNextGen is an effort to fill the gap, it may not be enough to satisfy IT’s needs.

Of the “hundreds of universities” claimed by IBM to focus on the mainframe, only 9 schools offer “at a minimum a certificate program on enterprise computing” .⁸   Only 1 of those schools is listed among the top 500 universities in the world.⁹ 

When comparing the mainframe TCO to other platforms in an apples-to-apples comparison, it is not done with another large centralized server such as an HP Superdome or HP NonStop server, rather it is typically against a large number of underutilized non-virtualized distributed servers.  Even though 50%-60%¹⁰ utilization is common in a virtualized x86 environment, IBM still uses 5%-10% utilization to justify the cost.  This makes it easy to then claim better heating, power, admin and facilities cost. 

Though IBM will claim that thousands of Linux servers can run on a mainframe, how can you independently validate these claims with objective evidence since IBM does not publish industry standard benchmarks on the mainframe?