IBM has quietly cut mainframe prices for years. With the economy in the dumps and x86-based chip vendors bringing out new high end processors, however, IBM can’t be coy about price cutting or risk losing deals to lower cost x86-based platforms.
The price cutting suddenly became quite public when The Register, a UK publication, ran a piece on IBM slashing IFL prices in half in the response to Intel’s latest chip, Nehalem. The article provided a detailed cost analysis of the System z versus comparable workloads running on systems running the latest Intel chip.
The analysis concluded that for the same money, “a fully-engined System z BC with five 3.2 GHz engines would need to support the same number of Linux images as 48 Intel-based systems running a total of 384 Nehalem cores running at 2.93 GHz.” At a minimum, that would translate into 384 Linux images.
The writer implies that the System z price/performance can’t match the newest generation of x86-based servers when running Linux workloads. I have not seen an IBM response to the article, but the writer’s cost analysis falls into a common trap. He compares initial acquisition costs and throws in some five-year support charges. That’s not nearly the whole cost story.
DancingDinosaur has covered this issue before, in March. The problem is that there are more costs to consider than just the hardware, software, and maintenance.
Three key cost issues were overlooked in The Register’s analysis:
- The System z is a more robust, secure environment. It was designed from the start to be industrial strength. You can aggregate several hundred Nehalem cores, but you will have to build the industrial strength attributes into the resulting system on your own if you hope to bullet-proof the system. That won’t be easy or fast and certainly won’t be cheap. You’ll have to add the cost of that to your calculus or add money to offset the increased risk, which will be significant.
- The people issue. Mainframe data center managers run lean shops. They don’t need a lot of people to handle the care and feeding of the System z. Shops running systems comprised of large numbers of commodity processors generally need considerably more people to keep those systems running at optimum levels. And while the cost of hardware and software goes down, the cost of people only goes up. If you want to run a cost-efficient data center, you have to minimize your people costs.
- The cost of energy. Admittedly, I don’t know what the energy consumption of 384 Nehalem processors comes to. The only figure I found puts the latest Nehalem delivering 10x the power while consuming 3x more energy than the previous generation. There were no comparisons to the System z, but somehow I doubt it is less than the System z running a comparable workload.
A recent survey by CA of executives at System z shops found 76% reporting that managing a large number of distributed servers in their data centers has become a cost issue. In addition, 67% noted that as the distributed infrastructure grows, the ability to run multiple applications on a single mainframe becomes more attractive. Hmmm.
So, don’t accept The Register’s cost analysis as the last word. Let’s see what the next System z processor rev brings.