Posted by: Matt Stansberry
By some estimates, since this past February when the z10 was introduced, Linux has accounted for more than half the total mainframe MIPS IBM has sold since the z10 was introduced last February. I asked a few experts to react to that statement:
Charles King of Pund-IT Research Hayward, Calif. is not surprised. “Linux has been on the mainframe for almost a decade,” King said. “For customers that already have mainframes — leveraging z/vm rather than buying mountains of x86 servers on a year over year basis been getting a lot of play and that’s where a lot of the additional numbers have been coming from.”
“IBM doesn’t make a lot of money selling Linux operating environments, but they do sell a lot of hardware and support services. Linux has been a boon for IBM and for the mainframe. I wonder if the IBM mainframe would continue to have the presence today if Linux wasn’t around.”
David Boyes of Sine Nomine Associates (the company responsible for porting Linux to Big Iron ten years ago) agreed, in email response:
Consider that [IBM] pretty much hands out IFLs like acid drops, and if you plan to do any serious DB/2 or Java work with z/OS, you pretty much have to have a zIIP or a zAAP in order to not drive your CA and IBM software bills through the roof. Since those specialty processors always run at full rated speed (as opposed to an increasing percentage of sub-capacity standard engines due to the loss of the Flex-ES machines for the low-end market and the high cost of z/OS software), the proportion of new IFL and zIIP/zAAP mips being sold dramatically outstrips the percentage of standard CPs being sold.
My take is that the amount of traditional z/OS processing (which is the critical demand driver for traditional mainframe MIPS) continues to decrease, but is being offset by new workload on z/VM and Linux, and additional MIPS on z/OS to handle processor intensive workload like Java, which is exactly where the zIIP and zAAP come into play. Now, the fact that we NEED that many MIPS for Java apps says more about the quality of the software and the tooling than the system it’s running on — there’s a lot of genuinely rotten Java code out there that simply assumes that processor resource is infinite, and you’ll just buy a bigger box if it doesn’t run efficiently — but a $400K price tag vs a $125K price tag for the specialty engines is a compelling argument to absorb new workload.
We contacted IBM for confirmation, but no comment so far.